Superconductor digital electronics: scalability and energy efficiency issues (Review Article)

Superconductor digital electronics using Josephson junctions as ultrafast switches and magnetic-flux encoding of information was proposed over 30 years ago as a sub-terahertz clock frequency alternative to semiconductor electronics based on complementary metal-oxide-semiconductor (CMOS) transistors....

Повний опис

Збережено в:
Бібліографічні деталі
Опубліковано в: :Физика низких температур
Дата:2016
Автор: Tolpygo, S.K.
Формат: Стаття
Мова:English
Опубліковано: Фізико-технічний інститут низьких температур ім. Б.І. Вєркіна НАН України 2016
Теми:
Онлайн доступ:https://nasplib.isofts.kiev.ua/handle/123456789/129109
Теги: Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Цитувати:Superconductor digital electronics: scalability and energy efficiency issues (Review Article) / Sergey K. Tolpygo // Физика низких температур. — 2016. — Т. 42, № 5. — С. 463-485. — Бібліогр.: 153 назв. — англ.

Репозитарії

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-129109
record_format dspace
spelling Tolpygo, S.K.
2018-01-16T13:33:06Z
2018-01-16T13:33:06Z
2016
Superconductor digital electronics: scalability and energy efficiency issues (Review Article) / Sergey K. Tolpygo // Физика низких температур. — 2016. — Т. 42, № 5. — С. 463-485. — Бібліогр.: 153 назв. — англ.
0132-6414
PACS: 85.25.–j, 85.25.Am, 85.25.Cp, 85.25.Hv
https://nasplib.isofts.kiev.ua/handle/123456789/129109
Superconductor digital electronics using Josephson junctions as ultrafast switches and magnetic-flux encoding of information was proposed over 30 years ago as a sub-terahertz clock frequency alternative to semiconductor electronics based on complementary metal-oxide-semiconductor (CMOS) transistors. Recently, interest in developing superconductor electronics has been renewed due to a search for energy saving solutions in applications related to high-performance computing. The current state of superconductor electronics and fabrication processes are reviewed in order to evaluate whether this electronics is scalable to a very large scale integration (VLSI) required to achieve computation complexities comparable to CMOS processors. A fully planarized process at MIT Lincoln Laboratory, perhaps the most advanced process developed so far for superconductor electronics, is used as an example. The process has nine superconducting layers: eight Nb wiring layers with the minimum feature size of 350 nm, and a thin superconducting layer for making compact high-kinetic-inductance bias inductors. All circuit layers are fully planarized using chemical mechanical planarization (CMP) of SiO₂ interlayer dielectric. The physical limitations imposed on the circuit density by Josephson junctions, circuit inductors, shunt and bias resistors, etc., are discussed. Energy dissipation in superconducting circuits is also reviewed in order to estimate whether this technology, which requires cryogenic refrigeration, can be energy efficient. Fabrication process development required for increasing the density of superconductor digital circuits by a factor of ten and achieving densities above 10⁷ Josephson junctions per cm² is described.
I am very grateful to all my colleagues at MIT Lincoln Laboratory who are involved with fabrication process development for superconductor electronics, especially to Vladimir Bolkhovsky and Scott Zarr, to Terry Weir and Alex Wynn for their part in device testing, and to Mark Gouker and Leonard Johnson for the discussions and management of the program. I would like to thank Vasili K. Semenov, Alex F. Kirichenko, Timur Filippov, Quentin Herr, Marc Manheimer, and D. Scott Holmes for useful discussions. My special thanks are to Daniel E. Oates for reading the manuscript and suggesting numerous improvements. This research is based upon work supported by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via Air Force Contract FA872105C0002. The views and conclusions contained herein are those of the author and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon.
en
Фізико-технічний інститут низьких температур ім. Б.І. Вєркіна НАН України
Физика низких температур
К 100-летию со дня рождения К.Б. Толпыго
Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
Article
published earlier
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
title Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
spellingShingle Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
Tolpygo, S.K.
К 100-летию со дня рождения К.Б. Толпыго
title_short Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
title_full Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
title_fullStr Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
title_full_unstemmed Superconductor digital electronics: scalability and energy efficiency issues (Review Article)
title_sort superconductor digital electronics: scalability and energy efficiency issues (review article)
author Tolpygo, S.K.
author_facet Tolpygo, S.K.
topic К 100-летию со дня рождения К.Б. Толпыго
topic_facet К 100-летию со дня рождения К.Б. Толпыго
publishDate 2016
language English
container_title Физика низких температур
publisher Фізико-технічний інститут низьких температур ім. Б.І. Вєркіна НАН України
format Article
description Superconductor digital electronics using Josephson junctions as ultrafast switches and magnetic-flux encoding of information was proposed over 30 years ago as a sub-terahertz clock frequency alternative to semiconductor electronics based on complementary metal-oxide-semiconductor (CMOS) transistors. Recently, interest in developing superconductor electronics has been renewed due to a search for energy saving solutions in applications related to high-performance computing. The current state of superconductor electronics and fabrication processes are reviewed in order to evaluate whether this electronics is scalable to a very large scale integration (VLSI) required to achieve computation complexities comparable to CMOS processors. A fully planarized process at MIT Lincoln Laboratory, perhaps the most advanced process developed so far for superconductor electronics, is used as an example. The process has nine superconducting layers: eight Nb wiring layers with the minimum feature size of 350 nm, and a thin superconducting layer for making compact high-kinetic-inductance bias inductors. All circuit layers are fully planarized using chemical mechanical planarization (CMP) of SiO₂ interlayer dielectric. The physical limitations imposed on the circuit density by Josephson junctions, circuit inductors, shunt and bias resistors, etc., are discussed. Energy dissipation in superconducting circuits is also reviewed in order to estimate whether this technology, which requires cryogenic refrigeration, can be energy efficient. Fabrication process development required for increasing the density of superconductor digital circuits by a factor of ten and achieving densities above 10⁷ Josephson junctions per cm² is described.
issn 0132-6414
url https://nasplib.isofts.kiev.ua/handle/123456789/129109
citation_txt Superconductor digital electronics: scalability and energy efficiency issues (Review Article) / Sergey K. Tolpygo // Физика низких температур. — 2016. — Т. 42, № 5. — С. 463-485. — Бібліогр.: 153 назв. — англ.
work_keys_str_mv AT tolpygosk superconductordigitalelectronicsscalabilityandenergyefficiencyissuesreviewarticle
first_indexed 2025-11-26T02:05:59Z
last_indexed 2025-11-26T02:05:59Z
_version_ 1850607931809071104
fulltext Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5, pp. 463–485 Superconductor digital electronics: scalability and energy efficiency issues (Review Article) Sergey K. Tolpygo Lincoln Laboratory, Massachusetts Institute of Technology, Lexington, MA 02420, USA E-mail: sergey.tolpygo@ll.mit.edu Received February 10, 2016, published online March 23, 2016 Superconductor digital electronics using Josephson junctions as ultrafast switches and magnetic-flux encod- ing of information was proposed over 30 years ago as a sub-terahertz clock frequency alternative to semiconduc- tor electronics based on complementary metal–oxide–semiconductor (CMOS) transistors. Recently, interest in developing superconductor electronics has been renewed due to a search for energy saving solutions in applica- tions related to high-performance computing. The current state of superconductor electronics and fabrication processes are reviewed in order to evaluate whether this electronics is scalable to a very large scale integration (VLSI) required to achieve computation complexities comparable to CMOS processors. A fully planarized pro- cess at MIT Lincoln Laboratory, perhaps the most advanced process developed so far for superconductor elec- tronics, is used as an example. The process has nine superconducting layers: eight Nb wiring layers with the minimum feature size of 350 nm, and a thin superconducting layer for making compact high-kinetic-inductance bias inductors. All circuit layers are fully planarized using chemical mechanical planarization (CMP) of SiO2 in- terlayer dielectric. The physical limitations imposed on the circuit density by Josephson junctions, circuit induc- tors, shunt and bias resistors, etc., are discussed. Energy dissipation in superconducting circuits is also reviewed in order to estimate whether this technology, which requires cryogenic refrigeration, can be energy efficient. Fabrication process development required for increasing the density of superconductor digital circuits by a factor of ten and achieving densities above 107 Josephson junctions per cm2 is described. PACS: 85.25.–j Superconducting devices; 85.25.Am Superconducting device characterization, design, and modeling; 85.25.Cp Josephson devices; 85.25.Hv Superconducting logic elements and memory devices; microelectronic circuits. Keywords: AQFP, ERSFQ, integrated circuit fabrication, Josephson junctions, kinetic inductors, Nb/AlOx/Nb junctions, RQL, RSFQ, superconductor electronics, superconducting integrated circuit. Contents 1. Introduction .......................................................................................................................................... 464 2. Fabrication and scalability of SFQ circuits .......................................................................................... 466 2.1. SFQ electronics fabrication processes .......................................................................................... 466 3. Physical constraints on VLSI of SFQ electronics ................................................................................. 469 3.1. Josephson junctions ...................................................................................................................... 469 3.2. Statistical variations of Josephson junctions ................................................................................ 471 3.3. Limitations on scaling caused by bias currents ............................................................................ 472 3.4. Heating of resistors ...................................................................................................................... 473 3.5. Circuit inductors ........................................................................................................................... 473 4. Energy dissipation and efficiency of SCE ............................................................................................ 474 4.1. RSFQ-based and RQL circuits ..................................................................................................... 474 4.2. Adiabatic quantum flux parametron (AQFP) circuits ................................................................... 477 4.3. Summary ...................................................................................................................................... 477 5. Future work .......................................................................................................................................... 477 5.1. Energy-efficient computing .......................................................................................................... 477 © Sergey K. Tolpygo, 2016 Sergey K. Tolpygo 5.2. High-speed computing ................................................................................................................. 478 5.3. VLSI technology development ..................................................................................................... 478 5.3.1. Self-shunted, high-Jc junctions .......................................................................................... 479 5.3.2. Number of junction layers ................................................................................................. 480 5.3.3. Compact inductors ............................................................................................................. 480 5.3.4. JJs as inductors .................................................................................................................. 480 5.3.5. Phase shifters and pi-junctions ........................................................................................... 481 6. Conclusion ........................................................................................................................................... 481 References ................................................................................................................................................ 482 1. Introduction May 3, 2016 is the 100th anniversary of birth of Kirill Borisovich Tolpygo, a prominent theoretical physicist widely recognized for his contributions to condensed mat- ter physics, crystal lattice dynamics, physics of semicon- ductors and dielectrics, and also biophysics [1–5]. Among his many works on application of mathematical and quan- tum-mechanical methods to biological systems, a signifi- cant part was devoted to developing an understanding of the mechanisms of high-energy efficiency of living organ- isms, in particular the mechanisms of chemical energy conversion into mechanical energy in muscles and muscle contraction. In 1978 Tolpygo proposed a mechanism of muscle contraction that results from a sequential transfer of proton excitation, a proton exciton, along a chain of hydro- gen bonds between two biopolymers, an actin–myosin pair [6,7]. A pulling force is produced due to lowering the ex- cited proton energy and shortening the bond length [8]. The initial excitation is provided by hydrolysis of an ade- nosine-triphosphate (ATP) molecule. A high-energy effi- ciency of muscle contraction is explained in this model as due to energy recycling — the energy remaining in a hydro- gen bond after a microscopic displacement of the polymers is transferred to a neighboring bond along the chain, and so on [9]. The original model was further developed by Tolpygo and his collaborators in a series of work; see [10–12] and references therein. It is likely that the idea of nearly complete energy recycling in muscles of living organisms can also be applied to explaining a high-energy efficiency of information processing by a human brain, the most en- ergy efficient computer created so far. The energy efficiency of electronics, in particular comput- ers, has become a very important problem due to an exponen- tial growth of energy consumption by computational and in- ternet-related systems: supercomputers, data centers, personal computers, etc., which is expected to reach ~ 15% of the total energy consumption in the world in the very near future. Any increase in energy efficiency of electronic systems, or of any area of human activity, would provide tremendous economic and environmental benefits by reducing global warming, achieving sustainable economic development, and protecting the environment. All these topics were of great interest and importance to K.B. Tolpygo, who lectured and published on them profoundly in the later part of his life. Conventional digital electronics is based on comple- mentary metal–oxide–semiconductor (CMOS) transistor technology where information is encoded by the voltage state of a field effect transistor (FET). The energy dissipa- tion is caused by charging and discharging of the circuit interconnects and gate capacitors of FETs and the gate current leakage in the “OFF” state of transistors. The charging energy is not recycled. Since the invention of the first integrated circuits in the 1960s, semiconductor digital electronics has demonstrated a nearly exponential growth of the integration scale and circuit complexity. The number of transistors per chip has grown by more than eight orders of magnitude, reaching over 1·109 (1B) in modern proces- sors and over 20B in field programmable gate arrays (FPGA). At the same time the size of transistors, their gate length, has shrunk from tens of microns down to 14 nm and continues to decrease. Although the gate length has been approaching the physical limits, there is little doubt that semiconductor industry will continue to pack more transis- tors per chip by using three-dimensional (3D) integration and other approaches for at least another decade. The pro- gress comes at a higher and higher cost, and the main hurdle is energy dissipation. It has reached ~ 100 W/cm2, a factor of 10× higher than the heat density of electric hot plates and induction burners, and a factor of 1000× higher than the solar energy density. Energy dissipation limits the clock frequency of processors to ~ 4 GHz and determines the amount of the so-called “dead silicon,” transistors which are not powered at any given time in order to prevent the chip temperature from exceeding the thermal limit. In the area of high-performance computing, the main interest is in advancing supercomputers from the current PFLOPS-scale (1015 FLOPS, Floating-Point Operations per Second) to exascale computing, corresponding to 1018 FLOPS and beyond [13]. A survey of the top 10 su- percomputers [14,15] gives their power consumption in 2015 at ~ 0.1 GW, with the most powerful supercomputer in 2015, Chinese Tianhe-2 (33.9 PFLOPS), consuming about 17.8 MW for operation and another 6.4 MW for cooling. A linear projection from this performance to a 1 EFLOPS (1000 PFLOPS) supercomputer using the same technology gives about 800 MW consumption, the output of an aver- age utility power plant. Current efforts in energy-efficiency improvements of supercomputers target a reduction of this figure to about 20 MW by about 2020 [16]. 464 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues Superconductor electronics (SCE) utilizing Josephson junctions (JJs) as switching devices was historically con- sidered for applications in high-performance computing mainly due to a potential for much higher clock rates (up to a factor of 50× higher) than those offered by the CMOS technology at that time [17]. Recently, superconductor digital electronics has been re-evaluated as having a poten- tial for energy-efficient computing, and energy consump- tion budgets and other technology requirements have been formulated [16]. A major research program, Cryogenic Computing Complexity (C3), was started in the US in Au- gust 2014 in order to develop the technology and demon- strate a prototype of a complete superconducting computer with 10-GHz target clock frequency [18]. Superconductor digital electronics operates at cryogenic temperatures, typically around 4.2 K, and there is currently no technology to enable operation of complex supercon- ducting circuits at significantly higher temperatures. There- fore, the energy required for cryogenic refrigeration must be included in the energy-efficiency calculations, which should include the energy dissipation in the circuits as well as all other sources of the heat load such as input/output data and power cables, thermal radiation, etc. Superconductor elec- tronics is also not self-sufficient. A superconducting circuit cannot operate without auxiliary semiconductor electronics such as power supplies, clock generators, output amplifiers, etc. Their energy consumption must also be included in the efficiency calculations. In order to become competitive with CMOS electronics, superconducting circuits must reach a very large scale of integration (VLSI) that would enable circuit functionalities and complexities required for computing. A result of the author’s survey of the Josephson junction count, the sim- plest measure of circuit complexity, in fully operational superconducting digital circuits, including also JJ-based memory and quantum annealing circuits, fabricated during the last 25 years is shown in Fig. 1. The data represent circuits made in the US and Japan by historically the most successful and currently available fabrication processes for SCE: by HYPRES 1 kA/cm2 and 4.5 kA/cm2 processes developed at HYPRES, Inc. [35–48]; by NEC-ISTEC 2.5-kA/cm2 standard process [49–64], by ISTEC-AIST 10-kA/cm2 advanced process (ADP and ADP2) [26,53,61,65–74], by the MIT Lincoln Laboratory 10-kA/cm2 process SFQ4ee [75,76], and by D- Wave Systems, Inc. [31–34]. Since any successful inte- grated circuit is usually a product of a joint work of circuit design and fabrication teams, Fig. 1 characterizes the state of affairs in both the superconducting circuit design and the circuit fabrication areas achieved during the last 25 years. A solid line in Fig. 1 shows an exponential growth with doubling the number of JJs in circuits every 4.5 years. This exponent is a factor of 3 smaller than in the exponen- tial growth demonstrated by CMOS industry during the same period by doubling the number of transistors every 18 months, often referred to as Moore’s law. A dashed line in Fig. 1 shows an exponential growth required for achiev- ing goals of the C3 program, doubling the number of JJs per circuit every year. At present, superconducting digital circuits have about five orders of magnitude lower integration scale than the typical CMOS circuits. E.g., the largest demonstrated Single Flux Quantum (SFQ) circuits have only about 105 Joseph- son junctions [68,75,76] whereas CMOS circuits routinely have over 1010 transistors. Assuming that all progress in CMOS industry stops right now and superconductor elec- tronics will be capable of sustaining the pace of doubling the number of JJs every year, it will take more than 16.5 years to catch up with the complexity of current CMOS circuits. Several causes of this gigantic disparity have been cited: insufficient funding; lack of profit-driven investments in SCE; immaturity of the fabrication processes and of inte- grated circuit design tools, etc. If these were the real causes, the corresponding solutions would be trivial: increase fund- ing; develop circuit design tools; use the modern design and fabrication tools; and improve the fabrication process. In the past there have been a few studies predicting the pace of SCE technology development. For instance, in [77] circuits with 1M (106) JJs were envisioned by 2005, with 10M JJs in 2008, and PFLOPS-scale superconducting computing in 2011. But no one can see the future and obviously none of these predictions turned out to be correct. Fig. 1. The total number of Josephson junctions in fully- operational superconducting integrated circuits reported in jour- nal publications and conference proceedings. The circuits were made by the following fabrication processes: HYPRES 1 kA/cm2 [19], HYPRES 4.5 kA/cm2 [20–22], NEC-ISTEC 2.5 kA/cm2 standard process [23,24], ISTEC-AIST advanced processes ADP [25,26] and ADP2 [27], MIT-LL SFQ4ee process [28–30], and D-Wave Systems process for quantum annealing processors [31–34]. The VLSI boundary corresponds roughly to 105 logic gates or ~106 JJs. A dashed line shows doubling the number of JJ in circuits every year. Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 465 Sergey K. Tolpygo In the present work we look a little deeper and, after a brief review of the operation principles and fabrication processes, look at the physical limitations of “classical” SCE and critically analyze whether this technology is en- ergy efficient and scalable to the integration levels required for high-performance computing. We also discuss potential approaches for reducing energy dissipation and increasing the integration scale. Superconducting qubits and numer- ous problems specific to their implementation in integrated circuits for quantum annealing and gate-based quantum computing will not be considered. 2. Fabrication and scalability of SFQ circuits Superconductor digital electronics utilizes magnetic- flux (fluxoid) quantization in superconducting loops to define and process bits of information. It is often called SFQ electronics. Nonhysteretic Josephson junctions are used as ultrafast switches. There are three main types of SFQ logic developed to date: Rapid Single Flux Quantum (RSFQ) logic/memory family [78,79] and its two new “en- ergy-efficient” versions — ERSFQ [80] and eSFQ [81] — differing from RSFQ only by the biasing schemes; Recip- rocal Quantum Logic (RQL) [82]; and Quantum Flux Parametron (QFP) logic [83,84] and its adiabatic (AQFP) implementation [85,86]. All superconducting digital circuits present a network of Josephson junctions interconnected by superconducting wires (inductors). DC bias currents are distributed using a network of bias resistors and a common voltage rail in RSFQ circuits or a network of bias inductors and JJs in ERSFQ and eSFQ. Multi-phase ac bias and clock signals in RQL and QFP circuits are distributed using a network of passive transmission lines (PTLs) and coupling transform- ers. The required switching properties of JJs are achieved using on-chip resistive shunting of hysteretic tunnel Jo- sephson junctions. So, making superconducting integrated circuits reduces to making large networks of JJs, inductors, resistors, and transmission lines. 2.1. SFQ electronics fabrication processes In the semiconductor industry the manufacturing processes are classified by the minimum feature size, which is the gate length of the FETs. Historically, the processes in SCE are classified by the critical current density, Jc of the Josephson junctions, e.g., 10-kA/cm2 process, which is the material property rather than the feature size characteristic. Although the area, A of the junctions is related to Jc as A = Ic/Jc, the connection with the process resolution capability is lost since JJs are not the smallest feature in circuits. Also, the ability to route various data and clock signals and interconnect logic cells grows with increasing the number of wiring layers avail- able. So this number and the minimum wiring feature size are also important characteristics. A review of fabrication pro- cesses up to 2004 can be found in [87]. At present, there are three advanced fabrication processes for SCE: at the National Institute of Advanced Industrial Science and Technology (AIST) in Japan; at D-Wave Sys- tems, Inc., using Cypress Semiconductor foundry in Bloom- ington, Minnesota; and at the MIT Lincoln Laboratory. The AIST process was reviewed in detail in [27,66]. It has two versions: advanced process (ADP) with 10 Nb layers; and ADP2 with 9 Nb layers. Their main limitation is the use of i-line photolithography, which limits the minimum feature size to ~ 0.8 μm, and the use of 3-inch wafers. The D-Wave process has 6 Nb layers, 0.25-μm minimum feature size, and is set on 200-mm wafers. There is no published de- scription of the process, but some information can be found in [31–34]. This proprietary process has been used mainly for making quantum annealing processors operat- ing at mK temperatures, which is a very different applica- tion than the digital SFQ electronics. In our recent work [28–30], we developed a fabrication process with eight Nb wiring layers and one layer of high- kinetic-inductance material for bias inductors, one layer of Nb/Al–AlOx/Nb Josephson junctions, and full planariza- tion, including the layer of Josephson junctions. So the total number of superconducting layers is nine. This pro- cess node was termed SFQ5ee, where “ee” denotes that the process is tuned for making energy efficient circuits for IARPA C3 Program [18]. Here we give a brief review of the MIT-LL fabrication process. The cross-section of the process is shown in Fig. 2. The target parameters of the layers as they appear in the pro- cessing and minimum feature sizes (critical dimensions) are given in Table 1. In comparison with the SFQ4ee pro- cess described in [28–30], a more advanced SFQ5ee node offers the following enhancements: i) the minimum linewidth and spacing for all metal lay- ers, but M0, M1, M5 and R5, is reduced to 0.35 and 0.5 μm, respectively; Fig. 2. Focused-ion-beam-made (FIB) cross-section of a wafer fabricated by the SFQ5ee process [30]. The labels of metal layers and vias are the same as in Table 1. The additional layers in SFQ5ee process with respect to the previous process node SFQ4ee are: a high-kinetic-inductance layer under M0 and a lay- er of mΩ-range resistors between M4 and M5 layers. 466 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues ii) the minimum size of etched vias and their metal sur- round is reduced to 0.5 and 0.35 μm, respectively; iii) the sheet resistance of the resistor layer is increased to 6 Ω/sq by utilizing a nonsuperconducting MoNx film, offering a choice of either 2 Ω/sq or 6 Ω/sq planar resistors for JJ shunting and biasing; iv) an additional thin superconducting layer with high- kinetic inductance is added below the first Nb layer M0 in order to enable compact bias inductors; v) an additional resistive layer is added between Nb layers M4 and M5 in order to enable interlayer, sandwich- type resistors with resistance values in the mΩ range for minimizing magnetic flux trapping and releasing unwanted flux from logic cells. The process consists basically of three main modules: wir- ing layer module; JJ module; and resistor/kinetic-inductor module. The wiring layer module is shown in Fig. 3. All wiring layers are processed identically as follows: (a) Nb layer Mi deposition; (b) deep-UV photolithography; (c) high-density plasma etching; (d) photoresist dry/wet strip. Then metrology steps follow: scanning electron mi- croscopy (SEM) inspection, critical dimension (CD) and thickness measurements. Planarization of the etched metal layer is done by a chemical mechanical planarization (CMP), using the steps shown in Fig. 3: (e) deposition of a ~ 2.5× times thicker SiO2 over the patterned metal layer; (f) pol- ishing SiO2 to the required level, using a CMP tool Mirra from Applied Materials, Inc. This is followed by the meas- urements of the remaining dielectric thickness in 49 points on the wafer, using an elipsometer, and redeposition of SiO2, if needed, to achieve the target ILD thickness in Table 1. Then, the next photolithography is done on the flat surface of SiO2, Fig. 3(g), in order to etch contact holes through the dielectric to the layer Mi, steps (g)–(i) in Fig. 3. Finally, the next wiring layer Mi+1 is deposited. This sequence of steps is repeated as many times as the number of wiring layers. All metal layers used in the process (Nb, Al, Mo) are deposited on 200-mm Si wafers by dc magnetron sputter- ing using a multi-chamber cluster tool (Endura from Ap- plied Materials, Inc.) with base pressure of 10–8 Torr. SiO2 interlayer dielectric (ILD) is deposited at 150 °C, using a plasma enhanced chemical vapor deposition (PECVD) sys- Table 1. Critical dimensions and layer parameters of SFQ5ee process Physical layer Photolithography layer Material Thickness, nm Critical dimension Ic a or Rs b Feature, nm Space, nm L0 L0 MoNx 40±10 2000 500 0.5 C0 C0 SiO2 60±10 500 500 M0 M0 Nb 200±15 500 500 20 A0 I0 SiO2 200±30 500 500 20 M1 M1 Nb 200±15 500 500 20 A1 I1 SiO2 200±30 500 500 20 M2 M1 Nb 200±15 350 500 20 A2 I2 SiO2 200±30 500 500 20 M3 M3 Nb 200±15 350 500 20 A3 I3 SiO2 200±30 500 500 20 M4 M4 Nb 200±15 350 500 20 A4 I4 SiO2 200±30 800 800 20 M5 M5 Nb 135±15 700 700 20 J5 J5 AlOx/Nb 170±15 700 1000 100 c A5a I5 anodic oxide d 40±2 700 700 A5b I5 SiO2 170±15 700 700 20 R5 R5 Mo 40±5 500 500 2±0.3 A5c C5 SiO2 70±5 500 500 20 M6 M6 Nb 200±15 350 500 20 A6 I6 SiO2 200±30 700 700 20 M7 M7 Nb 200±15 350 500 20 A7 I7 SiO2 200±30 1000 1000 n/a M8 M8 Au/Pt/Ti 250±30 2000 2000 n/a a Ic is the minimum critical current of metal lines or contact holes (vias) of the critical dimensions (CD) in mA. b Rs is the sheet resistance of resistor layer R5 in Ω/sq. c Josephson critical current density of Nb/AlOx–Al/Nb trilayer in μA/μm2. d Mixed anodic oxide (AlNb)Ox formed by anodization of Al/Nb bilayer of JJ bottom electrode. SiO2 was deposited by plasma enhanced chemical vapor deposition (PECVD) at 150 °C. Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 467 Sergey K. Tolpygo tem Sequel from Novellus (Lam Research Corporation). Thickness uniformity of the deposited oxide is σ = 2%, where σ is standard deviation (normalized to the mean val- ue). Photolithography is done using a Canon FPA-3000 EX4 stepper with 248 nm exposure wavelength, UV5 photoresist, and AR3 bottom antireflection coating. Etching of all metal and dielectric layers is done in a Centura etch cluster (Ap- plied Materials, Inc.), using Cl2/Ar-based chemistry for met- als and CHF3-based chemistry for dielectrics. Etched vias I0, I1, etc. are filled by Nb of the following metal layer. Josephson junction fabrication was described in detail in [28] and the JJ module is shown in Fig. 4. The Nb/AlOx–Al/Nb trilayer process developed in [88] has been the most successful process for making Josephson tunnel junctions and is used in our work. It consists of a Nb base-electrode deposition (150 nm) followed by in-situ Al deposition (8 nm) and oxidation in pure oxygen at 8 mTorr to achieve the aluminum oxide, AlOx, thickness required for 100 μA/μm2 Josephson critical current density. A Nb counter-electrode completes the trilayer sandwich. After etching the counter-electrode to define the Josephson junc- tions, the surface of the tunnel barrier is exposed. To pre- vent the barrier degradation around the perimeter of the junctions, anodic oxidation is used to form a ~ 50 nm oxide layer on all exposed surfaces, Fig. 4(d). This oxide protects the junctions and allows further processing steps: (e) pho- tolithography; and (f) etching of the bottom electrode in order to define Nb wiring layer M5 interconnecting the JJs and connecting them to the bottom layers, Fig. 4. Then, the etched structures are planarized to the level of the tops of JJs as shown by steps (g) and (h). In order to form resistively-shunted JJs, the following resistor process module is used, Fig. 5. A similar module is used to process the very first layer in the process stack-up, Fig. 3. Processing module of the wiring layers: (a) deposition of a wiring layer Mi; (b) deep-UV photolithography; (c) Nb etching in high-density plasma; (d) photoresist dry/wet strip; (e) plasma enhanced chemical vapor deposition (PECVD) of SiO2 interlayer dielectric for planarization; (f) chemical mechanical planarization (CMP) of the interlayer dielectric (ILD) to the required thickness; (g) deep-UV photolithography of the interlayer dielectric layer Ii; (h) SiO2 etching; (i) photoresist dry/wet strip and surface clean- ing; (j) deposition of the next Nb wiring layer Mi+1. This next Nb layer fills in the etched contact holes in the ILD, thus forming su- perconducting vias between Nb layers. The sequence of steps is repeated as many times as required by the number of wiring layers. Fig. 4. Josephson junction fabrication module: (a) Nb/AlOx–Al/Nb trilayer deposition over the patterned SiO2 layer. The base elec- trode of the trilayer fills the etched contact holes, making I4 vias to the bottom Nb layer, M4. (b) deep-UV photolithography of the counter electrode to form a junction etch mask; (c) junction etch- ing, stopping on AlOx/Al layer; (d) anodization; (e) photolitho- graphy of the base electrode layer; (f) etching of the base elec- trode, forming wiring layer M5; (g) deposition of a thick SiO2 for planarization; (h) CMP to the level of the Josephson junctions to expose their top surface. Fig. 5. Resistor/kinetic-inductor process module: (a) resistor dep- osition; (b) resistor photolithography, high-density plasma etch- ing and photoresist striping; (c) SiO2 layer deposition, 70 nm thickness; (d) photolithography and etching of contact holes to JJ and resistor C5, photolithography and etching of contact holes to the base electrode of JJs, M5; (e) Nb wiring layer deposition, M6. Nb layer M6 and layers above it are processed using the wiring module in Fig. 3. The resistor minimum length is determined by the minimum spacing s between superconducting wires and con- tact holes surround sr shown in (e). 468 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues the layer of kinetic inductors, L0, Fig. 1. A resistor layer (Mo or MoNx) is deposited on the planarized surface. After the photolithography, resistors are selectively etched in high-density plasma, stopping on Nb and SiO2. A thin, with ~70 nm thickness, SiO2 layer is deposited on top to isolate the resistors. Contact holes to the top and bottom electrodes of the junctions and to the resistors are etched. Nb wiring layer M6 is deposited. This M6 layer and layers above it are processed using the wiring module shown in Fig. 3. The final process cross-section is shown in Fig. 2 and a zoom-in of a cross-section through the junction is shown in Fig. 6. The full 9-superconductor-layer process has more than 400 processing steps. A perceived simplicity and alleged low cost of the fab- rication process were historically cited as one of the main advantages of SFQ electronics [77,79]. At the time of RSFQ introduction in the US in 1991, the only commercial fabrication process for SCE had only three superconduct- ing niobium layers for interconnecting Josephson junc- tions, a minimum feature size of 3.5 μm, used Nb/AlOx/Nb Josephson junctions, and 3-inch wafers [19]. At that time, these process features corresponded to about Intel’s pro- cess used to make the semiconductor processors in 1982, and so the process was about 10 years behind, see Table 2. The first and very simple RSFQ circuits containing tens of JJs showed great promise and were setting the clock speed records. So, from this starting point, it was tempting to forecast a fantastic growth in future. Since none of the original RSFQ technology proponents was involved with or experienced in integrated circuit technology and manufac- turing, it was natural to assign circuit failures to immaturity of the fabrication processes [77] and suggest that their major improvement would be a simple task requiring only very modest investments and second-hand tools from retiring nodes of CMOS manufacturing lines [17]. After 25 years of the SFQ fabrication-technology development, it is clear that these assessments were incorrect. The minimum linewidth of SCE processes has shrunk down to 0.25 μm, the number of superconducting wiring layers has increased from 3 to 9, and the wafer size has increased to 200 mm. So, the features of the currently available SCE processes match and exceed the Intel’s process used to manufacture Pentium II processors in 1997, see Table 2. However, no SFQ-based computers or digital circuits with complexities comparable to any of the CPUs shown in Table 2 have emerged. 3. Physical constraints on VLSI of SFQ electronics By comparing the features of the SCE fabrication pro- cesses described above with CMOS processes given in Table 2, one could expect that superconducting circuits with a similar integration scale, similar number of JJs in a few million range, and of similar functionality should be possible to fabricate if the appropriately designed circuits become available. Below we examine this expectation by accounting for the specifics of SFQ circuits. 3.1. Josephson junctions Firstly, we estimate the maximum possible density of unshunted junctions in SFQ circuits. The total area occu- pied by a circular junction in M5 (base electrode) or M6 (top wiring) planes is 2 1/2 2 /2( ) [ / ( / ,) 2]J c cA r sr s I J sr s= π + + = π π + + (1) where cJ , r, sr and s are the Josephson critical current den- sity, the junction radius, base electrode and top wire (M6) surround of the junction, and spacing to the next object, respectively. Then, using s/2 = sr = 0.25 μm from Table 1, we plotted in Fig. 7 the maximum density of unshunted Fig. 6. FIB-made cross-section of a 1-μm Josephson junction. In this particular case, the contact C5 is larger than the junction J5, so the M6 wire overhangs the junction. The opposite situation when C5 is smaller than J5 was shown in Fig. 5. Anodic oxide layer formed on the junction sidewalls and the surface of M5 by anodization is clearly visible. Table 2. Semiconductor processors and fabrication processes Processor Transistor count Min linewidth, μm No. of metal layers Year introduced Chip area, mm2 Intel 80186 55k 3.0 2 1982 60 Intel 80286 134k 1.5 2 1982 49 Intel 80386 275k 1.5 2 1985 104 Intel 80486 1.2M 1.0 3 1989 173 Pentium 3.1M 0.8 3 1993 294 Pentium Pro 5.5M 0.5 4 1995 307 Pentium II 7.5M 0.35 4 1997 195 Pentium III 9.5M 0.25 5 1999 128 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 469 Sergey K. Tolpygo JJs, /J Jn k A= as a function of their critical current cI , as- suming a 100% area coverage, k = 1. In the range of the critical currents typically used, from 100 μA to 300 μA, nJ is from about 15M to 30M JJs per cm2. This coverage, of course, is impossible to achieve in a circuit because JJs need to be interconnected to other circuit components. At a more realistic k = 0.5, Jn is comparable to the density of transistors in the processors in Table 2, but three orders of magnitude less than the typical density of modern CMOS transistors. This gap cannot be closed even if the junction technology is pushed to the ultimate values: cJ increased 25 folds to ~ 2.5 mA/μm2 [92]; sr and s/2 reduced to 100 nm; and the number of JJ layers increased to two, still giving only Jn ~ 5·108 JJ/cm2. SFQ circuits utilize nonhysteretic junctions. In the ex- isting technology this is achieved by resistive shunting of the tunnel junctions as shown in Fig. 5. The top view of a resistively shunted junction (RSJ) is shown in Fig. 8. The total area of the RSJ includes the junction area JA , resistor area, and the area of vias and wires connecting the JJ and the resistor. This area includes also the overlap between wires M5 and M6 and the contact holes and the junction by amount sr, a process parameter given in the design rules document. Below we estimate the total RSJ area in order to estimate the maximum circuit density. SFQ circuits utilize RSJs with critical damping, i.e., with McCumber–Stewart [89,90] parameter cβ = 2 02 / 1,c nI R C= π Φ ≈ where 0 /2h eΦ ≡ is the flux quantum, C is the junction capacitance, nR is the damping resistance assumed to be a parallel combination of the junction inter- nal resistance R and the shunt resistance sR , see inset in Fig. 9. The inductance Ls associated with the supercon- ducting connections to the shunt and of the shunt itself makes damping frequency-dependent, which is usually neglected in SFQ circuit design. The internal resistance is usually approximated by a piecewise function: for voltage ,ssg N gR R R V V= = γ < (2a) at ,N gR R V V= ≥ (2b) where gV = 2Δ/e is the gap voltage in the symmetrical tun- nel junction, Δ is the energy gap in the electrodes, RN is the normal state tunnel resistance, sgR is the subgap resistance, Fig. 7. The maximum density of Josephson junctions in SFQ circuits as a function of the average Ic of the SFQ cells. Resistive- ly shunted and unshunted JJs in the current technology node SFQ5ee (Jc = 0.1 mA/ μm2, sr = s/2 = 0.25 μm) are shown by the solid lines; self-shunted JJs in a hypothetical technology node with Jc = 2.5 mA/μm2 and sr = s/2 = 0.1 μm are shown by the dashed line. A 100% area coverage is as sumed, k = 1. Fig. 8. Top view of a resistively-shunted JJ showing JJ counter electrode J5, JJ bottom electrode M5, resistor R5, contact holes C5 and I5, and wiring layer M6 providing connections between the JJ and the resistor. All metal wires M5 and M6 must overlap (surround) the JJ and all contact holes by some amount, sr ac- cording to the process design rules. Fig. 9. Current-voltage characteristics of resistively shunted Jo- sephson junctions in the MIT-LL fabrication process SFQ5ee with Jc = 100 μA/μm2. Data for 1.6-μm-diameter junctions with three different shunt resistors Rs are shown: Rs = 1.6 Ω (1); Rs = = 3.73 Ω (2); Rs = 5.4 Ω (3), corresponding to Vc ≡ IcRn values of 0.30, 0.69, and 0.96 mV, respectively. These values in turn corre- spond to βc = 0.2, 1, and 2, respectively. The top I–V curve 4 is for the unshunted JJ. Inset shows the circuit diagram. The shunt inductance associated with the current path from the JJ along the resistor to the I5 via and back to the JJ along the M5 electrode, Ls ~ 1 pH, is usually neglected. However, it causes an internal Ls-C resonance with the junction capacitance, corresponding to a step at ~ 0.5 mV in curve 2. 470 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues and γ is the temperature- and JJ-quality-dependent coeffi- cient, γ > 1. Then, neglecting Ls, the damping resistance at low voltages gV V< is /( ) n N s N sR R R R R=γ γ + . The typical current-voltage characteristics of JJs in the MIT Lincoln Laboratory process using Nb/Al–AlOx/Nb tunnel junctions with Josephson critical current density of 10 kA/cm2 (100 μA/μm2) are shown in Fig. 9 for the unshunted junction (curve 4) and junctions of the same size with three different values of the shunt resistor corresponding the characteristic voltage c c nV I R≡ of 0.30 mV (curve 1), 0.69 mV (curve 2), and 0.96 mV (curve 3), and to cβ = 0.2, 1, and 2, respectively. At the specific capacitance value of 70 fF/μm2 given in the SFQ5ee process design rules, the characteristic voltage of the junctions at cβ = 1 is c c nV I R≡ ≈ 686 μV. At cJ = 100 μA/μm2 used in the process, ,sg sR R>> ,NR γ ≈ 10 in (2), and its contribution can be neglected in the estimates of the shunt resistor here. Then, the shunting resistor value is simply ( )/ 686/ ,s c c cR V J A I= = where sR is in ohms and cI in μA. The area of a resistor sq /sR R l w= with the minimum linewidth w and length l depends on the sheet resistance of the material used, sqR and other process parameters, see [30], 2 sq sq/ if ( )/ 2 / ,R s sA w R R R R s sr w= ≥ + (3a) 2 sq sq 2 / if / 2 )/( ) ( ,R s sA s sr R R R R s sr w= + < + (3b) because the resistor length cannot be made shorter than lmin = s + 2sr, about 1 μm in the current process node. At 2 / ,sqR sq= Ω the boundary corresponds to 4 .sR = Ω Therefore, all JJs with Ic ≥ 172 μA have shunts in the re- gime (3b). We need to add the area of two C5 and one I5 vias with surround, which is approximately 23 ( )2w sr+ in the regime (3a) and 2 sq 2 2 ( ) ( )(2 2 /) sw sr s sr w sr R R+ + + + in the regime (3b), and account for the spacing to the next fea- ture, where w is the minimum size of features, see Table 1. Then the total area of an RSJ becomes 1/2 2 sq / /2 [( ) ] ( /) ( )RSJ c c c cA I J sr s w w s V I R=π π + + + + + 2 3 2 ,( )w sr s+ + + (4a) for sq / 2[ ( )] /c cI w s sr V R< + ≈ 172 μA, and 1/2 2[( / /2) ] ( )2RSJ c cA I J sr s s sr=π π + + + + × 2 sq2 2 / 2[ ( ) ] ( ) c cs sr s sr I R V w sr s+ + + + +× + + sq( )[ ( 2 2 2 /) ]c cw sr s s s sr I R V+ + + + + (4b) for cI ≥ 172 μA. The maximum possible density of RSJs 1/RSJ RSJn A= following from (4) is plotted in Fig. 7, bottom curve. It is significantly lower than the maximum density of unshunted junctions. The maximum density is about 8.3M RSJs per cm2 and nearly independent of the critical current of JJs in the range from ~ 70 μA to ~ 175 μA. The RSJ area in this range is ARSJ ~ 12 μm2. The maximum density of RSJs in SFQ circuits can be estimated as / RSJk A by using in (4) the most frequently encountered, or the average, critical current cI and the area filling factor k ~ 0.5. Inspection of all RSFQ, ERSFQ, etc. cells in [77–81,91] shows that cI ≈ 175 μA. It is a result of selecting cI ≈ 100 μA as the minimum val- ue used in the cells, based on the maximum acceptable bit error rate. Since the junction must be connected to induc- tors, the RSJ area coverage of 25% to 50% is more realis- tic, reducing the maximum circuit density to about 2M to 4M RSJs per cm2. 3.2. Statistical variations of Josephson junctions As was shown above, the maximum density of RSJs is nearly independent of the choice of cI and sufficient to place a few million of JJs on a 1-cm2 chip. However, it is important to check if they all can be yielded with critical currents within the required margins. From the circuit de- sign standpoint, statistical variations of JJ critical currents, as well as thermal and quantum noise, induce storage, de- cision, and timing errors and determine the bit error rate in SFQ circuits [93]. From the fabrication process standpoint this is related to the so-called parametric yield — the fabri- cation yield of devices with parameters lying within a giv- en range ±M with respect to the targeted mean value. The SFQ cells are designed to tolerate some relatively large deviations. For instance, all critical currents of the junc- tions can be changed by about ±30% if changed uniformly, or any single junction can deviated from the target by the ±30% if all other JJs are on target, etc. However, random deviations of all junctions and all inductors in the cells cause significant margin shrinkage. The typical bias mar- gins reported for the circuits shown in Fig. 1 are often less than ±10% especially at high clock frequencies. The statistics of JJ critical currents fabricated by our planarized process was studied in [28] and can be de- scribed as approximately Gaussian with the standard devia- tion depending strongly on the junction size. This depend- ence comes from the fluctuations of the junction area caused by photolithography and tunnel barrier transparen- cy fluctuations, both increasing with decreasing the junc- tion diameter. Statistical data in [28] can be converted into the dependence of the standard deviation of cI (normal- ized to the mean value) on the critical current (in μA) as 1/2 00.115 /( ),I c cI I Iσ = − (5) where the dimensional prefactor is in μA1/2 and 0I is the critical current cut-off — the process resolution characteristic corresponding to the critical current of the smallest resolvable junction. In the SFQ5ee process using 248-nm photolithogra- phy, this minimum size is about 250 nm and 0I ≈ 5 μA. Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 471 Sergey K. Tolpygo The probability that a JJ critical current is within the circuit margins ±M is given by the error function (erf 2 .)/ Ip M= σ For an N-junction circuit, the probabil- ity (yield Y) that all JJs are within ±M range is Y = pN. Then, the fabrication-yield-limited maximum number of JJs in a circuit is max minln /ln erf( ) [ ( 2 ,)]/ IN Y M= σ (6) where minY is the minimum acceptable circuit yield. This dependence at minY = 50% and 90% is plotted in Fig. 10 for three different values of the circuit margins M: 15%, 10% and 5%. Also shown is the maximum number of RSJ which can be placed on a chip with area chA ch /RSJ RSJN kA A= (7) at fill factor k = 0.5 and chA = 1 cm2 and 2 cm2. At critical currents smaller than the point of intersection of (6) and (7) the maximum number of JJs in the opera- tional circuits is determined by the cI spreads, acceptable circuit yield, and the circuit margins. At higher critical currents, this number is limited only by the junction area and can be increased by increasing the size of the chip. It can be seen that at cI ≈ 175 μA and above, as was chosen in RSFQ originally, the size of the circuit (JJ count) is not limited by the parametric yield of the current process even for poorly designed circuits with ±5% margins, and the circuit parametric yield above 50% can be reached even on large-area chips ~ 2 cm2 with about 10M RSJs. However, the current desire for energy efficiency requires reducing cI well below this number, see Sec. 4. Below cI = 50 μA, a value popular in many RQL and AQFP designs and used in [16] for calculating SCE electronics power budgets, the circuit complexity will be limited by the fabrication process unless circuits with very wide margin can be designed. It is author’s experience, however, that the practical yield of complex circuits is much lower than the paramet- ric yield following from the assumed normality of the pa- rameter distribution used in our estimates here. In practice, the circuit yield is determined by defects and outlier devic- es, i.e., devices in the far tails of the distribution. The probability of outliers increases with decreasing the junc- tion size. The distribution is often skewed and the tails are usually non-Gaussian. Often they can be described by the Weilbull statistics with a simple exponential decay. A more detailed description of this subject is beyond the scope of this work and will be presented elsewhere. How- ever, the message of this is that decreasing cI below ~ 75 μA is expected to compromise the circuit yield in the current technology node. 3.3. Limitations on scaling caused by bias currents RSFQ-based circuits, including ERSFQ and eSFQ, use a parallel dc biasing of JJs from a common voltage rail. The typical bias current is 0.7Ic. The total bias current Itot grows proportionally to the number of JJs and can be estimated as tot 0.7 ,RSJ cI N I= (8) where RSJN is the number of JJs in the circuit, giving about 122 A per 1M JJs. It is clear that such large currents cannot be supplied to and handled by thin-film supercon- ducting layers at 4 K. Large bias currents also create large stray magnetic fields and cause magnetic flux trapping and circuit margin degradation. The maximum current which has successfully been delivered to the largest operational RSFQ circuit is ~ 3 A [69], with a substantial margin deg- radation of some of its sub-circuits. This total-bias-current limitation caused the saturation in the number of JJs in RSFQ circuits at ~ 12 thousand (12k) JJs, which can be seen in Fig. 1 during a 10-year period from ~ 2004 to 2014. All circuits with JJ count over 20k shown in Fig. 1 used various ac serial-biasing schemes. RSFQ-based circuits or any circuits with parallel bias- ing are not scalable beyond about 20k to 30k JJs. To miti- gate this problem, serial biasing of RSFQ circuits was pro- posed a long time ago [94] and demonstrated in relatively simple circuits in [95–97]. Serial basing, also known as current recycling, requires breaking a circuit into m isolat- ed islands and recycling the return current from the ground planes of one island to bias in series the next island, thus reducing the total current by a factor of m. The current drawn by each island must be equal, and the input and out- put currents must not add currents to the serially biased circuits. This is achieved by transferring microwave clock Fig. 10. The maximum number of Josephson junctions which can be yielded with a required probability (circuit yield) Y by the current fabrication process SFQ5ee and with all junctions having the critical current within the circuit margins ±M, as given by (6) and (5). Dashed lines and solid lines correspond to Y = 0.5 and Y = 0.9, respectively. Also shown is the maximum number of resistively shunted JJs kAch/ARSJ which can be placed on a chip with 1 cm2 and 2 cm2 area at 50% area coverage, k = 0.5, and Jc = 100 μA/μm2. 472 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues and SFQ data between the islands through transformers using driver-receiver pairs [98]. For an m-bit processor, a natural recycling scheme would be between the m individ- ual bits. Since the number of JJs per island should be kept at ~ 10k level in order to keep the total current below 2 A, the number of islands in a 2M-junction circuit becomes > 100, requiring lots of inter-island interface circuitry with its own biasing. This complicates design and decreases the circuit density. Presently, there are no readily available and proven solutions for making VLSI SFQ circuits with current recycling, and many technical problems remain to be solved. RQL and QFP circuits use multiphase ac currents for clock and bias, so the total current only weakly increases with .RSJN The technical difficulties with ac biasing lie in the proper distribution of these high-frequency currents along PTLs to each junction and the negative effect on the circuit density caused by the PTLs and coupling transformers. 3.4. Heating of resistors Many SFQ circuits use resistors and resistive dividers for distributing dc bias currents and matching rf impedance. These resistors are in direct contact with superconducting wires on M6 layer through C5 vias. Heat dissipated in the resistors increases the local temperature and decreases the critical current of wires and JJs, and in extreme cases can turn them into the normal state. This should be mitigated by a proper sizing of resistors and wires, and proper spacing of high-current-carrying resistors from other circuit elements sensitive to temperature increase. The amount of heat dissipated in a thin-film resistor with sheet resistance ,sR width w, and length l per unit time is 2 diss sq / ,Q I R l w= (9) where I is the current. In the steady state, this amount of heat power is balanced by the heat conduction through the circuit layers into the substrate and into liquid helium (or a cryocooler) cooling the chip surfaces. The amount of heat that can be removed due to conduction can be estimated as cond th/ ,RQ A T R= ∆ (10) where 2RA lw= is the resistor surface area, thR is the effective thermal resistance, and ΔT is the resistor tempera- ture increase. Equating (9) and (10) we get 2 2 sq th ( ) / 2 ,T I R R w∆ = (11) i.e., the resistor temperature increases inversely with the resistor width squared. There is a maximum temperature increase maxT∆ above which Nb wires in contact with the resistor will transition into the normal state at a given cur- rent and given bath temperature. This sets the maximum value of the current in the resistor as 1/2 max max sq th2 /[ ( )] .I w T R R= ∆ (12) The maximum current one can supply through the resistor without turning the contacting Nb wires into the normal state is inversely proportional to the square root of the sheet resistance. So the desire to increase sqR in order to minimize the resistor area, see (3a), is at odds with the re- sistor heat handling capabilities. Since heating of SFQ cir- cuits is highly undesirable, (11) and (12) put strong re- strictions on the minimum width of resistors that can be used or the currents they can handle, strongly impacting the integration scale. The thermal resistance at both interfaces of the resistor is not exactly known. For a metal surface in contact with liquid helium at 4.2 K, the typical thermal resistance thR is about 1 K⋅µm2/µW. Using this value of thR , 2 /sq,sR = Ω and max Nb cT T∆ = – 4.2 K = 5 K we estimated the maxi- mum current per unit width of the resistor as max /I w= = 2.2 mA/µm. Simple measurements of I–V characteristics of Nb wires in contact with Mo resistors of different width done in this work give max /I w= 1.3 mA/μm. This value translates into the effective thermal resistance of thR ≈ ≈ 3 K⋅µm2/µW (3·10–6 K⋅m2/W) for the circuit resistors buried deep into the multilayered structure with multiple interfaces between Nb and SiO2 layers (Fig. 1). Resistor heating makes it almost impossible to achieve very large integration scale in circuits with parallel biasing. For instance, distributing 244 A bias current in a circuit with 2M JJs, considered in Sec. 2.3, would require a total width of bias resistors of ~ 20 cm. 3.5. Circuit inductors Yet another constraint comes from the circuit inductors, since every JJ in an SFQ circuit is connected to an induc- tor. By inspecting the published RSFQ and RQL cells and circuits, we note that the average superconducting loop with JJs in the designs has a dimensionless parameter 02 / ~ 2,L cI Lβ = π Φ where L is the inductance. We denote this average value as .Lβ Each inductor occupies area ( )( )/ ,LA L w s= + where , w and s are the inductance per unit length, inductor linewidth, and spacing between the in- ductors, respectively. The maximum inductor density grows linearly with the average critical current of SFQ cells as 0[2 / ],( )L c L Ln I m k w s= π Φ + β (13) where Lm is the number of physical layers of inductors, k is the filling factor. In order to minimize cross-talk between in- ductors, stripline configuration is used predominantly. This requires three superconducting layers: one signal layer and two ground planes. So, a process with eights superconducting layers may have up to three completely independent layers of inductors. The smallest inductor linewidth in our pro- cess is 0.35 μm and spacing is 0.5 μm, so w+s = 0.85 μm. At this linewidth, the typical stripline inductance per unit length is  ≈ 0.6 pH/μm [29]. Since, on average, each RSJ requires an inductor, the maximum circuit density can be es- timated by equating RSJn and Ln . Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 473 Sergey K. Tolpygo The plot of (13) is shown in Fig. 11, along with RSJn and Jn as a function of the average critical current cI , for a few values of Lβ and ,Lm and k = 0.5. (It should be noted that some fraction of the area is occupied by vias providing connections between the inductors and the junc- tions, which may reduce the actual value of k below 0.5.) We can see that circuits with two layers of inductors can reach the maximum of about 3M components (both RSJs and inductors) per 1-cm2 chip if cI is larger than about 35 μA. Circuits with just one layer of inductors may still reach the same complexity if cI is larger than ~ 75 μA. The latter approach has been taken in designing ac-biased shift registers [75], and circuits with densities over 0.6M RSJs per cm2 have been demonstrated. Recently these shift registers have been redesigned and fabricated at MIT-LL by the SFQ5ee process. Operational circuits with 1.3M RSJs per cm2 density and over 65000 RSJs have been demonstrated [99]; see Fig. 12. Testing of shift registers with 144000 RSJs is the work in progress. At cI lower than the intersection point of the ( )L cn I and ( )RSJ cn I dependences, the circuit complexity is lim- ited by the number of inductors and, at higher cI , by the number of junctions. Circuits with cI larger than about 125 μA are not limited by inductors at all, only by the area occupied by RSJs. It is also clear from Fig. 11 that increasing Lm beyond 2 by increasing the number of superconducting layers in the process will not increase the circuit density, because it is lim- ited by the density of RSJs. However, more layers may add more flexibility in routing the clock, bias, and data paths. It is interesting to note that, if the resistive shunts could be eliminated by using self-shunted JJs with the same cI and the similarly tight parameter spreads as the shunted tunnel junctions, the circuit complexities can reach about 10M (JJ/inductors) per 1-cm2 chip by choosing cI in the range from ~120 μA to 190 μA. 4. Energy dissipation and efficiency of SCE 4.1. RSFQ-based and RQL circuits RSFQ logic was proposed more than 30 years ago as a replacement of Josephson latching logic used by IBM in its early attempt to build a Josephson-junction-based comput- er in the 1970s–early 1980s. In RSFQ logic/memory the information is encoded by the presence (logic 1) or ab- sence (logic 0) of a single flux quantum Φ0 ≡ h/2e in a logic cell and is transferred between the cells in the form of picosecond-wide SFQ voltage pulses. An SFQ pulse is a voltage pulse generated across a Josephson junction when the phase difference across the junction flips by 2π as a result of some external perturbation, e.g., a current pulse. The second Josephson relation [100] / (2 / )d dt e Vϕ =  guarantees that these SFQ pulses have a quantized area equal to a single flux quantum 0( )V t dt =Φ∫ ≈ 2.07 mV∙ps or 2.07 pH·mA. The process of SFQ pulse generation can be also viewed as a passage of a flux quantum through the junction. Accordingly, if a junction is embedded into a superconducting loop, such a passage also changes (reduc- es or increases) the flux through the loop by Φ0. A com- plete description of SFQ logic was given in [79] and the cell library can be found in [91]. Each cell has separate data inputs and outputs, and clock lines. SFQ pulses encod- ing data and clock are distributed on two different net- works of JTLs and PTLs. In the time domain, logic “1” is Fig. 11. The maximum density of circuit inductors in the SFQ5ee process as a function of the average critical current in SFQ cells cI , assuming the average Lβ = 2 and Lm = 2 (curve 1), where Lm is the number of independent layers of inductors. Also shown are: Lβ = 3, Lm = 2 (curve 2); Lβ = 2, Lm = 1 (curve 3); and Lβ = 3, Lm = 1 (curve 4); the density of shunted and unshunted junctions RSJn and Jn . Inductance per unit length of 0.6 pH/μm [29] was used, and the area fill factor k = 0.5 was assumed. Fig. 12. An example of the unit cell (bit) of the ac-biased shift registers [99]. The cell dimensions are 20 μm × 15 μm, the RSJ density is 1.3·106 RSJs per cm2. A 16363-bit fully operational circuit has 65536 RSJs. The minimum linewidth, mL, and Ic used were 0.4 μm, 1, and 125 μA, respectively. 474 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues encoded by the arrival of a data SFQ pulse (on the data input) between two clock pulses (on the clock input), whereas logic “0” corresponds to no data pulse between the clock pulses. To provide reliable switching by an SFQ pulse and set the direction of SFQ pulse/flux propagation, almost all JJs in the circuits are current-biased at a value 0.7 ,b cI I≈ us- ing a network of either bias resistors and a common volt- age rail (as in RSFQ) or a network of inductors and cur- rent-limiting junctions (as in ERSFQ, eSFQ), or their combinations. The average dynamic-power dissipation in an SFQ cir- cuit utilizing Josephson junction switching (RSFQ, ERSFQ, eRSFQ, RQL) can be estimated as cold cl swP Nf E=α , (14) where N is the number of Josephson junctions in the cir- cuit, clf is the clock frequency, α is the activity coeffi- cient, the fraction of JJs switching during the clock period, swE is the average energy loss per switching. It is well known that a resistively capacitively shunted junction (RCSJ) has a potential energy described by a tilted washboard potential: 1– co( ) ,( s – )JE E iϕ = ϕ ϕ (15) where 0 / 2( )J cE I= Φ π is the Josephson energy and /b ci I I= is the normalized bias current. When switched by an SFQ pulse, the junction goes from one potential minimum of (15) to another separated by a 2π phase difference. The energy difference between the two minima is 2 .JE iE∆ = π Since the junction is critically damped or overdamped, all of this energy is dissipated in the resistor, and there is no energy left for recycling and increasing the efficiency of the information processing. Hence, the average energy loss per switch is sw 0 ,bE I= Φ (16) where bI is the average bias current. This energy should not be confused terminologically with the switch- ing energy — the energy required to flip the junction’s phase by 2π. The latter equals to the height of the potential barrier between the two adjacent minima of (15) and can be made arbitrarily small by increasing the bias current. Using the typical value 0.7 ,b cI I= following from the circuit speed optimization, and assuming a random mix of “ones” and “zeros” (α ≈ 0.5), the average power loss in an RSFQ-type circuit is cold cl 00.35 .cP Nf I= Φ (17) The RQL circuits use four-phase ac currents for JJ bias- ing and circuit clock [82]. A reciprocal pair of SFQ pulses (positive and negative) during the clock period encodes “1” and no pulses encode “0”. This encoding increases the energy loss by a factor of two, and for a random mix of “ones” and “zeroes” results in α = 1 in (14). Also, according to [82], the switching of JJs by an ac-bias current occurs at a current smaller than Ic. As a result, the energy dissipation in RQL circuits with random data can be approximated by cold cl 0( )1/3 ,cP Nf I= Φ (18) where cI is the weighted average of the critical current of the junctions [82]. The difference between (18) and the RSFQ-based case (17) is completely negligible. Inspection of all RSFQ cells in [77–79,91] shows that the average critical current cI = 0.175 mA. This gives the average energy loss swE = 2.5·10–19 J or 0.25 aJ per JJ switching in RSFQ, ERSFQ, and eSFQ circuits. This is a factor of 6·103 times larger energy loss than the Landauer’s minimum energy-per-bit requirement for irre- versible computing, ln 2Bk T [101] at T = 4.2 K; Bk is the Boltzmann constant. It is important to stress that the mini- mum critical current used in the SFQ cells is typically a factor of two lower than the average as a wide range of JJs values is used in the cells. However, the minimum critical current sets the maximum acceptable bit error rate [77,79] and typically is about 0.1 mA. Changing the minimum cI value by a factor of m would automatically change the cI and the swE by the same factor. Another important quantity is the energy loss per a sin- gle-bit operation (SBOP). It is different from swE be- cause any logic gate (Boolean) operation requires switch- ing of multiple JJs, so sw ,SBOP SBOPE N E= (19) where SBOPN is the average number of JJ switches re- quired. This number in RSFQ and RQL cells is about 10. For instance, OR gate has 12 JJs, XOR gate has 9 JJs, AND gate 11 JJs, etc. [77,79,91]. Recall that three SFQ pulses (3 switches) are required just to encode “1” and two switches to encode “0”. So, SBOPE ≈ 10 swE ≈ 2.5 aJ. In order to compare the energy efficiency of the cryo- genic electronics with room-temperature electronics we need to account for the energy loss associated with cryocooling. Removing coldP from the chip at 4.2 K re- quires a cryocooler (a heat machine) consuming a much larger power, hotP at 300 K, given by hot cold cold hot cold cold( /) ,( )/P P P T T T= ε= − η (20) where idε=ηε is the energy efficiency of the cryocooler and cold hot cold cold hot( ) / /id T T T T Tε = − ≈ is the efficiency of an ideal thermal machine (Carnot efficiency) and 1 η< is a nonideality factor. Depending on the cryocooler size and type, this factor changes from ~ 0.02η for the small-scale (~1.5 W at 4.2 K) pulse-tube coolers to ~ 0.20 for the larg- est-scale Linde helium liquefiers with ~ 400-kW wall power requirements, see, e.g., Table 1 in [16]. The inverse efficiencies of these two types of cryocoolers, sm1/ε = 3520 and 1/ lε = 352, are used hereafter for all power consump- tion estimates as the upper and the lower boundaries. Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 475 Sergey K. Tolpygo In order to determine which technology is more energy efficient, CMOS or SFQ, we need to compare the power loss in two circuits performing a similar function or a similar amount of information processing. The performance and power loss in the CMOS-based processors can easily be measured, are well known and can be found in [102]. The power dissipation is typically below 140 W. For instance, Intel’s quad-core Core i7-4790 Haswell CPU using 22-nm technology node has 88-W power dissipation at clf = 4 GHz, performance of about 200 GFLOPS, and N = 1.4·109 tran- sistors [103]. Unfortunately, superconducting processors of comparable complexity do not exist, and the achieved inte- gration scale differs by 5 orders of magnitude. So, to make a comparison, we need to make some assumptions. Below, we provide a few estimates of the upper and lower bounds on the power consumption in VLSI SFQ circuits. Firstly, we note that the number of JJs in RSFQ and RQL logic gates and memory cells is comparable or even larger than in CMOS logic gates and memory. Secondly, superconducting processor architectures that are being de- veloped use algorithms developed for CMOS computers and emulate their architectures, using adders, multipliers, etc., see, e.g., [47,63–67,71–74,104]. It is reasonable to assume then that a superconducting processor with N junc- tions operating at clf will be processing about the same amount of information (perform the same logic and memory functions) as a CMOS-based processor with N transistors clocking at the same frequency. Then, using clf = 4 GHz, N = 1.4·109 JJs, and swE = 2.5·10–19 J estimated above for cI = 0.175 mA, we get from (17) and (18) coldP = 0.71 W, see Table 3. This is a low power com- pared with about 100 W power consumed by the CMOS chip operating at room temperatures. However, depending on the cryocooler used, the total power consumption by our hypothetical SFQ circuit is from hotP ≈ 250 W to 2.5 kW, a factor of ~ 3× to 30× larg- er than in the CMOS processor of the same complexity. This, perhaps, is a rather surprising result for many read- ers. Of course, a fraction of transistors in the CMOS pro- cessor is sleeping at any given moment in order to prevent overheating. This was not taken into account in our esti- mate. The same approach can also be implemented in SFQ electronics. Because this result may look too pessimistic, we decid- ed to check it by comparing the power requirements per GFLOPS in CMOS and SFQ implementations. The only existing data for operational bit-serial, single-precision (32-bit) floating-point adders (FPA) and floating-point multipliers (FPM) made in RSFQ technology were reported in [69]. They have, respectively, 16830 JJs and 18766 JJs. (For a comparison, a 32-bit adder requires about 3000 transis- tors.) The measured performance is shown in Table 4 along with the data for an i7-4790 Haswell CPU. We used the above cited thermal efficiencies of the cryocooler to get the lower and upper bounds of power consumption and computation efficiency (in GFLOP per joule) at room tem- perature. Basically, we get the same result as above: RSFQ-based FPA and FMP are 2 to 20 times less efficient than the off-shelf CPU, Table 4. The numbers reported in [69] correspond to the RSFQ circuits using bias resistors with significant static power dissipation. If this dissipation is eliminated by using, e.g., ERSFQ approach, the compu- tation efficiency may improve by a factor of ~10, making the SFQ circuits competitive if implemented in a very- large-scale system, but still losing in efficiency if used in a small-scale system. If the average critical current in SFQ circuits can be re- duced by a factor of five, to 35 μA, somehow keeping the acceptable bit error rate determined by the smallest junc- tions in the circuit, the total energy dissipation in complex SFQ circuits with account for refrigeration may become somewhat lower than in their CMOS counterparts. Note, however, that we have not accounted for any power con- sumption associated with auxiliary electronics, thermal radiation, and heat conduction via cryogenic cables. A more optimistic estimate of computation efficiency of SFQ processors can be obtained by estimating naively the energy consumption per PFLOP, using the energy per single- bit operation SBOPE estimated above. Depending on the circuit architecture, a 32-bit floating-point operation requires about 1500 single-bit operations (for adders) and somewhat more for multipliers. Then, the lower bound on the energy loss per FLOP at 4.2 K is FLOPE ≈ 2000 SBOPE ≈ ≈ 5·10–15 J, or 5 J per PFLOP, giving a computation effi- ciency of ~ 0.2 PFLOP/J at 4.2 K or from about 57 to 570 GFLOP/J at 300 K. This is from 25 to 250 times better than CMOS, see Table 4. Unfortunately, there are currently no architectural solutions to realize this ultimate computa- tion efficiency because superconductor electronics does not have compact and efficient memory, whereas memory is abundant in semiconductor computers. So, it follows from the estimates above that cryogenic computational systems based on the SFQ logic versions utiliz- ing JJ switching and emulating CMOS architectures will like- ly be faster than the CMOS-based but not necessarily more energy efficient. This has a very simple reason — require- Table 3. Power dissipation in CMOS and hypothetical VLSI SFQ processors Unit fcl, GHz N (109JJs or transistors) Pcold at 4.2 K, W Phot, W lower bound η = 0.2 Phot, W upper bound η = 0.02 SFQ 4 1.4 0.71 250 2500 i7-4790 4 1.4 – 88 88 476 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues ments for switching elements used in any computing sys- tem are basically the same and independent of the operat- ing temperature. These are requirements of reliability, driva- bility, and communications: fast switching times ~ 1 ps, immunity to the thermal noise and parameter spreads (low bit error rates), and ability to drive other parts of the sys- tem and communicate with them. These requirements set the minimum ratio sw / ,BE k T typically ~ 104, for the switches used in a computer, where T is its operating tem- perature. Then, of two computers operating with the same sw / BE k T ratio and having similar architectures, the one requiring refrigeration to coldT may become more energy efficient than a computer operating at hotT only if the re- frigerator is nearly ideal, cold hot1 / ,T Tη≥ − which is not possible at cryogenic temperatures. 4.2. Adiabatic quantum flux parametron (AQFP) circuits Among the superconducting digital technologies devel- oped to date only AQFP circuits can be truly energy effi- cient. AQFP is the same QFP invented by E. Goto more than 30 years ago, see [83,84] and references therein, but operated in a slow (adiabatic) regime [85,86]. Instead of using JJ switching to move fluxons, as in RSFQ and RQL, the information in QFP is encoded by a fluxon location in a double-well potential, in the left (“0”) or the right (”1”) well, and moved by adiabatically varying the shape of the potential, using ac bias currents. The measured energy dis- sipation at 5 GHz operation is extremely low, 0~ 0.1 cI Φ per bit [147] and can be further reduced. QFP is very simi- lar in the operation principle to a much older device — parametric quantron [148,149]. Theoretically, these types of parametric devices can provide for the lowest energy consumption in computations [150]. Because parallel pipelining is very natural for AQFP, processors with higher computational efficiency than ERSFQ, RSFQ, and RQL can be designed [151]. Accord- ing to estimates in [151], the computational energy effi- ciency of some algorithms implemented in AQFP could be 7 orders of magnitude higher, with account for refrigera- tion, than of CMOS circuits. 4.3. Summary A very inquisitive reader may ask why our assessment of energy efficiency of RSFQ and RQL technologies for high-performance computing differs from the one made in [16], which indicated that RQL might be able to meet the efficiency requirement. The difference comes from simple arithmetic. In the simplistic estimate of the power required to compute 1 PFLOPS, see Eq. (1) in [16], the junction activity factor α was double-counted: the first time implic- itly in the estimate of energy per switch in the RQL 01/3( ) ,cI Φ which already includes α = 1/2; and the se- cond time explicitly in the power estimate based on the number of gates per FLOPS. Also, in this estimate, the minimum critical current of JJs of 25 μA was used instead of the weighted average cI entering (16)–(19), which is typically a factor of two larger than the minimum cI . As a result the power per PFLOPS was probably underestimated by about a factor of four, and the energy efficiency was overestimated by the same amount. The use of the mini- mum cI in energy-consumption estimates is very typical for reviews of SCE; see, for example, [77,79,118]. It im- plies that a superconducting computer (circuit) can be built from the identical junctions having the minimum critical cur- rent and the minimum area. This could suffice for an order-of- magnitude estimate, but it is not a valid assumption. 5. Future work The choice of the most promising directions of future work strongly depends on the strategic goals one wants to accomplish. In a limited funding environment, setting the priorities and optimizing the strategy should be done thor- oughly because “no one can have his cake and eat it too”, and diverting time, efforts, and funding toward one area may leave the other one starving, and so on. The author’s selection is given below and should not be construed in any form as funding recommendations. 5.1. Energy-efficient computing If the primary goal is energy efficiency, then a clear winner among the existing and relatively mature technolo- gies is AQFP, because its energy consumption, including refrigeration can be made a few orders of magnitude lower than in the existing room-temperature technologies. How- ever, the clock speed of truly energy-efficient AQFP cir- cuits is limited to ~ 7 GHz. Due to the use of multiple transformers and ac power, the cell area is currently large and the integration density is low. The limits on the inte- gration scale are the same as for other types of SCE as de- scribed in the previous sections. Unfortunately, there is Table 4. Power dissipation and performance of RSFQ [69] and CMOS floating-point processors Unit fcl, GHz Throughput (GFLOPS) Power at 4.2 K, mW Power dissipation at room-T, W Efficiency at room-T (GFLOP/J) FPA 58 2.23 4.92 1.7 to 17a) 0.13 to 1.3 FPM 59 2.36 5.76 2.0 to 20a) 0.12 to 1.2 i7-4790 4 200 – 88 2.27 a) Using the inverse efficiencies of 352 (η = 0.2) and 3520 (η = 0.02). Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 477 Sergey K. Tolpygo currently no work on AQFP circuits for high-performance computing anywhere except Japan. Interestingly, however, most of the control circuitry in D-Wave quantum annealing processors operating at tens of mK is based on QFPs due to their ultralow power dissipation. Even higher energy efficiency is promised by reversible (or almost reversible) computing. Ideas of reversible com- puting with superconducting circuits were discussed in [107–110]. Simple circuits, shift registers, with energy-per- bit near the Landauer’s thermodynamic limit ln 2Bk T have already been demonstrated [110]. Larger circuits with rich- er functionalities need to be developed and fabricated. Un- fortunately, the progress in the area of superconducting reversible computing has been slow due to lack of funding. As was shown above, the energy efficiency of SFQ processors utilizing junction switching and CMOS archi- tectures, i.e., designed by replacing CMOS logic cells by SFQ logic cells, is marginal, and the energy saving may not be sufficient to warrant the effort. Therefore, the obvi- ous way of making SFQ processors more energy efficient is to employ different information processing solutions and architectures that would require significantly fewer JJs than transistors to implement, and would not require exter- nal memories. These ideas have been discussed to some extent but require practical development. For instance, V.K. Semenov in [105] argued that RSFQ blocks should be implemented in a way that preserves their inherent logic and memory functions, and makes use of the simplicity and record-high speed of SFQ T-flip-flops [106] instead of replicating CMOS logic cells. Interesting to note, the RSFQ technology inventors stated in their original com- prehensive review [79] that “a universal von-Neumann- type computer is probably the worst device for implemen- tation using the RSFQ (or any other superfast) technolo- gy”, and explain why. Somehow this message and the idea that RSFQ blocks with their logic/memory functions are cellular automata or finite-state machines rather than CMOS-type logic gates were forgotten in the course of the last 25 years. Unfortunately, the only idea that is being actively worked on is the most trivial one — reducing the average critical current cI in SFQ cells. This reduction has a clear limit ~ 50 μA, below which the circuit density, bit error rate, and circuit yield will be substantially compromised, and hence cannot be a long-term strategy. On this road, RQL has currently a big advantage because of the serial biasing. For instance, SFQ circuits with the largest JJ count per chip demonstrated so far have been the RQL shift reg- isters [76] made by the MIT-LL SFQ4ee process [28]. It does not mean that RQL is a better technology however, only that its problems are in a different area. RSFQ-based circuits are clearly behind in JJ count due to the parallel biasing and associated problems. Therefore, the prime goals should be in developing serial-biasing (current recy- cling) solutions suitable for multimillion-JJ circuits. With- out solving this problem, RSFQ-based circuits, in the au- thor’s opinion, have no future in high-performance compu- ting or other applications requiring VLSI. 5.2. High-speed computing On the other hand, if the primary goal is the computa- tion speed, high clock frequency, and energy dissipation is secondary, then the technology selection and the develop- ment priorities are very different. RSFQ is clearly the fast- est digital technology developed so far, and capable of reaching ~ 70 GHz clock frequencies in the current tech- nology node and over 100 GHz if the Jc is increased to 0.5 mA/μm2 and beyond. Therefore, development of cur- rent recycling for VLSI circuits becomes the priority num- ber one. Since resistive biasing has no place in supercon- ductor VLSI due to heating, ERSFQ becomes a clear winner if its inductive biasing approach can be scaled up to the VLSI. RQL and AQFP circuits will probably lose the speed competition because of the multiphase ac biasing. Design development priorities in this case are also dif- ferent. Instead of reducing the critical current cI , it should be increased and optimized for junctions with high- er Jc, increased to ~ 0.5 mA/μm2 and above, perhaps self- shunted JJs, which will likely have larger parameter spreads at the same sizes as the current tunnel junctions with Jc = 100 μA/μm2. As priority number two, I would rate the development of new architectural solutions that do not copy the standard CMOS. It is the author’s opinion that SCE electronics can- not win or even be competitive if it mimics CMOS, be- cause of the five orders of magnitude difference in integra- tion scale. Number three on my list would be the development of fast and compact JJ-based memories that do not use mag- netic materials and magnetic tunnel junctions discussed in [111–113], a development that may take many years. Jo- sephson random access memories (RAM) have been demonstrated in older process nodes with large features and low RAM densities [23,114,115,152]. They should be improved and implemented in the advanced process nodes using more advanced materials and smaller features. 5.3. VLSI technology development The main advantage of RSFQ-based processors is that they can run much faster (likely 25 times) than the 4 GHz offered by CMOS because energy dissipation on the chip is significantly reduced, see (17), (18), and moved instead to a much larger cryocooling system at room temperature. So, our hypothetical chip with 1.4B JJs can run at 100 GHz and dissipate only about 18 W at 4.2 K. The typical power that can be removed from the chip without raising its tem- perature more than 1 K in liquid He is ~ 1 W/cm2. So this chip can be cooled if its active area is larger than ~ 18 cm2. This should be an easy task because a SCE chip with 1.4B RSJs would have an area of 700 cm2 at the present maxi- 478 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues mum density of 2M RSJs per cm2. A chip of this area is difficult to imagine and would be impossible to manufac- ture, so a superconducting multichip module (MCM) with about 200 of 2-cm2 chips could be dreamed of as an equi- valent. The required MCM technology with SFQ interchip communication data rates exceeding 100 GHz has been demonstrated; see [116,117] and references therein. This example demonstrates that the development of VLSI technology for SCE trumps all the priorities men- tioned above. Without solving the scalability problem, de- velopment of SFQ processors has no merit. Note that fab- rication technology development and implementation of new materials and processes are usually much more expen- sive and time consuming than circuit design because the former requires expensive and sophisticated processing equipment whereas the latter requires only good ideas and engineers with CMOS-based computers. This simple truth was mainly ignored during the last 25 years. Many ideas of what could be done have been floated around. I briefly review them in order to rate their impact on increasing the circuit density and feasibility of imple- mentation. To be specific, let us set an increase in the cir- cuit density by a factor of ten, i.e., achieving 2·107 cm–2 density of JJs and inductors, as the primary near-term goal. (The author simply cannot imagine a 200-chip MCM for our 1.4B-JJ processor, but can imagine a 20-chip MCM.) 5.3.1. Self-shunted, high-Jc junctions Getting rid of resistive shunts would have the largest im- pact on the circuit density, as is clear from Fig. 7, and would shorten the fabrication process by eliminating the resistor module, Fig. 5. This requires replacing the hysteretic SIS tun- nel junctions with nonhysteretic junctions. In this context, we would like to clear up one of the misconceptions in this area, which appeared in [77,119,153] and crept into many other publications. It is an incorrect assertion that all tunnel junc- tions become self-shunted at high Jc, e.g., ~ 100 kA/cm2 in the case of Nb-based junctions. This misconception is a result of a simple-minded application of the McCumber–Stewart [89,90] parameter 2 02 / ,c c nI R Cβ = π Φ derived for a linear shunt resistor, to tunnel junctions with nonlinear, voltage- dependent, damping (2). Then, using NR as a damping resistance at all voltages, noting that the expression can be rewritten as 2 0( ) ( )2 / ,c c N s cI R C Jβ = π Φ where sC is the junction specific capacitance, and that c NI R is the elec- trode material property ~ Δ/e, it may appear that 1 cβ ≤ can be obtained by mere increasing the cJ . This is, of course, incorrect because there are not enough electronic states at V < Vg to provide damping ( )sg NR R>> , and switching back into V = 0 state is hysteretic. Therefore, self-shunting can be induced only by increasing the density of states in the gap or in the barrier allowing for a substantial ohmic conduction (junction “leakage”) at V < Vg. This can be achieved by introducing defects in the tunnel barrier, e.g., pin-holes and oxygen vacancies [120–124], or by using junction barriers with direct conduction such as normal met- als, doped semiconductors, etc. However, high-quality tunnel junctions, e.g., using Nb2O5 and AlNx barriers instead of AlOx, remain highly hysteretic at Jc values much larger than 100 kA/cm2 [125–127]. This makes Nb/Al–AlNx/Nb tunnel junctions unattractive for VLSI applications requiring self- shunted JJs, although they are perfectly good junctions for SIS detector applications. Among devices with direct conduction, relatively high values of Vc required for digital applications have been demonstrated by Nb-based junctions with amorphous Si barriers doped by various impurities creating deep levels near the middle of the band gap, like Nb, W, etc.; see [128], references therein, [129–131], and many older works, e.g., [132–134]. From the author’s point of view, their only advantage is that, at large levels of doping (~ 10%), self-shunting and nonhysteretic behavior is ob- tained. The other properties are rather drawbacks. Their Vcs are inferior to Nb/AlOx/Nb junctions at the same Jc, and the temperature dependence of the critical current is stronger than in SIS junctions. The self-shunting in these devices is likely due to inelastic and resonance tunneling via multiple localized sates, percolation paths, within the barrier [128,135]. As a result of this unusual conduction mecha- nism, the I–V characteristics are nonlinear and deviate from the RCSJ model; the devices also have unusually high spe- cific capacitance, which is larger than in SIS devices. The circuit clock speeds based on these devices are also re- duced by ~ 30% with respect to the same circuits made with SIS junctions [130]. It is not clear whether -Siα -doped devices are reproducible at submicron dimensions required for their implementation in VLSI circuits, because of the percolation-type current transport and possible fluctuations in the number of localized centers and resonant paths. A good uniformity of -Si:Nbα barriers in voltage standard applications was demonstrated using JJs with areas of hun- dreds of square micrometers. These data cannot be project- ed onto submicron-scale devices. Therefore, the most important task is to evaluate the critical current spreads of -Siα -doped JJs with submicron dimensions at critical current densities in the range from 0.1 mA/μm2 to about 1 mA/μm2 by using the modern pro- cessing tools similar to the one used in [28–30]. Somehow this has not been done despite these junctions being around for over 35 years. (Sperry Corporation was trying to com- mercialize superconducting devices and circuits using Nb and NbN junctions with doped -Siα , -Geα , and Si–Ge barriers in the 1980s.) The same comment applies to high-Jc Nb/AlOx–Al/Nb junctions. Their subgap transport is due to multiple Andreev reflections via defects, most likely oxygen vacancies, in the barrier. So they could be viewed as ap- proaching the metal-insulator transition from the opposite side than -Siα -doped barriers. There has been a claim that, at high-Jc, the transport mechanism in Nb/AlOx–Al/Nb junctions becomes universal [136], so the junction-to- Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 479 Sergey K. Tolpygo junction variations could be small. However, the experi- mental basis for this is about 200 high-Jc JJs studied in [92,137], which is not sufficient to build a VLSI technolo- gy on. So, the near-term goals should be the evaluation of junctions with Jc = 0.5 mA/μm2, five times larger than it is in the SFQ5ee process, in order to measure their parameter spreads and self-shunting. One of the drawbacks of all compact self-shunted JJs is self-heating. Heating in the RSJs is negligible because the swE is dumped into the resistor whose area is much larger than the JJ area. In the self-shunted JJs the same energy dissipates inside the junction and creates nonequilibrium quasiparticles. Energy density and heating increase with increasing Jc. The maximum clock frequency of these de- vices may then be determined by the energy relaxation rate in order to prevent memory effects and time jitter associat- ed with the influence of one switching on the next and dif- ference in activity factors in different parts of the circuit. 5.3.2. Number of junction layers Increasing the number of independent JJ layers to two is beneficial and may increase the JJ density by a factor of 2×. The work in this direction is currently underway at AIST in Japan. A larger increase would likely have no proportional effect on the density because of the need for interconnecting and through vias reducing the available area. 5.3.3. Compact inductors The next in importance is the increase in inductor densi- ty Ln . This can be done by decreasing their area, increas- ing the inductance per unit length, and increasing the num- ber of layers, Lm . Increasing Lm above two will have little impact because of the vias and cross-talk problems. At small spacing between the inductors ~ 0.25 μm, the cross- talk between the inductors on the same layer or between the layers becomes significant. So the inductors must be isolated by the ground planes and vias between them, boxed into a coaxial-type configuration. This reduces the maximum density and defeats the purpose. Let us take as an example, the average inductor in the cells of the ac-biased shift register in [75,99] and shown in Fig. 12. Its value is about 6 pH, and in the current process it occupies the area of ~ 8.5 μm2. A factor of ten increase in Ln means that its area needs to be reduced to ~ 0.85 μm2. This can be achieved by using a thin layer of a high- kinetic-inductance material like MoNx or NbNx for signal inductors, similar to our SFQ5ee process for bias inductors [30]. It is well known that the kinetic inductance of thin superconducting films 2 0 /kL t=µ λ is much larger that their geometric inductance if tλ>> and 2 / ,t wλ >> where λ is magnetic field penetration depth, t and w are the film thickness and width, respectively. The kinetic-inductor layer can be placed in a close proximity to JJs, e.g., be- tween layers J5 and M6 instead of the layer of resistors, R5 in Fig. 2, since with self-shunted JJs resistive shunts will no longer be required. The Ic of this kinetic inductor should be a factor of 2× larger than Ic of JJs, or about 0.25 mA. Then, using a typical value of inductance per square of ~ 5 pH/sq, e.g., for a 60-nm-thick MoNx film, and the film width of 0.5 μm to guarantee the critical cur- rent, the kinetic inductor length will be 0.6 μm and the area, accounting for a 0.25-μm line spacing, will be about 0.5 μm2. Adding vias to the inductor will double the area. A compact via process has been demonstrated [138]. So one layer of kinetic inductors near the self-shunted JJs can increase the circuit density by a factor of ten, and two lay- ers by a factor of ~ 20×. The use of kinetic inductors does not help in reducing the area of inductive couplers and transformers in RQL and AQFP circuits, which depend on the geometric mutual inductance. The author has no readily available solutions for increasing the density of RQL and AQFP circuits by a factor of ten, except for reducing the minimum linewidth and spacing of inductors down to ~ 100 nm. 5.3.4. JJs as inductors Nonswitching Josephson junctions can be used as cir- cuit inductors. This idea is very old, but so far has been only implemented in superconducting persistent-current qubits [139–140] and their advanced versions [141], and in tunable rf filters [142]. A segment of Josephson transmis- sion line with JJs replacing all inductors would look like the one shown in Fig. 13. The area saving can only be achieved if a vertical stack of inductor junctions is used. In order to preserve the de- sign of all main SFQ cells, the minimum number of junc- tions in the stack should be three. Then, a quantizing in- ductor with a 2π total phase shift can be formed using two JL units shown in Fig. 13 (top panel). The phase across each JJ in the stack will be ~ /3,π far enough from /2π when the critical current is reached. We would like to term Fig. 13. A complementary JJ-inductor pair, a CLJJ pair, formed by a junction and a vertical stack of three nonswitching JJs, LJ. The ability to apply bias and signals at the point between the JJ and LJ is important and shown explicitly in the circuit diagrams in the top panel. Bottom panel: a segment of Josephson transmis- sion line (JTL) using CLJJ pairs, i.e., with all inductors replaced by series arrays of nonswitching Josephson junctions LJ. The switching junctions are J1, J2, etc., Ib is the bias current. 480 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 Superconductor digital electronics: scalability and energy efficiency issues a switching JJ and three stacked nonswitching inductor-JJs a complementary pair, or a CLJJ pair, as shown in Fig. 13. Accordingly, circuits built using this technology can be termed complementary-SFQ circuits or CSFQ circuits. Since there is no switching speed requirement for the inductor-junctions in the stack, the Vc of the junctions can be lower than of the switching JJs. Then, a much simpler technology of SNS junctions can be used to form the stack, where N is a normal metal compatible with the process, e.g., Al, Mo, Ti, etc. No doped -Siα or other fancy barriers are required. They could also be used, but are not necessary. The ability to apply bias and signal at the point between the JJ and LJ is important and shown explicitly in the cir- cuit diagrams. The fabrication process can be sketched briefly as shown in Fig. 14. The typical critical current of nonswitching JJs, cLI can be estimated from the typical value of ~ 2,Lβ using in- stead of magnetic inductance L the Josephson (kinetic) inductance 0 (3 / 2 ),J cLL I= Φ π which gives 3/ ),(cL LI = β or ~1.5 .cL cI I To satisfy our 10× density increase require- ment, the JL area should be less than 1 μm2, so the critical current density of nonswitching JJs should be larger than 100 μA/μm2. The process shown in Fig. 14 would be relatively easy to implement. However, the drawback of replacing thin- film inductors with JJ-inductors is that a very simple and highly reliable device — a narrow strip of high-kinetic- inductance material — is replaced by much more complex and less reliable devices using a multilayered stack of JJs. The JJs in the stack need to have (nearly) identical critical currents. This is a serious complication of the process and a potential impediment to the process reliability and yield. Moreover, implementation of CLJJ pairs does not solve the problem of miniaturization of inductive couplers and trans- formers in RQL and AQFP circuits. 5.3.5. Phase shifters and pi-junctions The use of phase shifters based on pi-junctions with fer- romagnetic barriers has been discussed and demonstrated in [143–145]. Despite their useful features and interesting physics, adding pi-junctions would have a negative impact on the circuit density, because the critical current density of pi- junctions is a factor of ~100 lower than in regular 0-junctions and their area accordingly is a 100× times larger. The ex- isting advanced fabrication processes have so many super- conducting layers that this function (phase shifting) could be easily realized by using a miniature rings and narrow wires to provide magnetic bias; see, e.g., [146]. 6. Conclusion In order to evaluate whether SFQ electronics is scalable to VLSI levels required for achieving computation com- plexities comparable to CMOS processors, we have re- viewed the current state of the fabrication technology for superconducting digital circuits based on single-flux- quantum encoding of information. We have described the limitations imposed on the circuit density by Josephson junctions, circuit inductors, shunt and bias resistors, pa- rameter spreads, etc. We have shown that the currently achievable maximum circuit density of resistively shunted Josephson junctions and inductors is about 2·106 cm–2, which is almost four orders of magnitude lower than the density of transistors in modern CMOS circuits. We have described the fabrication-process development required for increasing the density of SFQ digital circuits by a factor of ten, including self-shunted Josephson junc- tions, kinetic inductors, complementary JJ-inductor pairs (CLJJ) using Josephson inductance of stacked, nonswitching junctions, etc. Energy dissipation in superconducting cir- cuits has been also reviewed in order to estimate whether SFQ electronics, which requires cryogenic refrigeration, can be energy efficient in comparison with CMOS. We estimated that energy dissipation in SFQ circuits based on JJ switching, such as RSFQ, ERSFQ, RQL, and having comparable complexity to the modern CMOS processors will be comparable to that in CMOS processors when re- frigeration energy is taken into account. The energy efficiency can be improved if innovative circuit architectures could be developed, which use much fewer JJs than transistors for the same information pro- cessing and have minimum use of external memories. The most energy efficient technology is adiabatic quantum-flux parametron (AQFP), which can run at ~ 7 GHz clock fre- quency. RSFQ and its “energy efficient” versions remain the fastest digital superconductor technology capable of clock frequencies over 100 GHz, if energy efficiency is not Fig. 14. A process for making complementary LJ-JJ pairs, CLJJ pairs: regular JJs coupled to inductors formed by three stacked nonswitching JJs. After finishing the switching JJ planarization step, see Fig. 4(h), the resistor module in Fig. 5 can be abandoned because we plan to use self-shunted JJs. Then, a wiring layer M6 will be deposited and patterned (b). After M6 planarization by the dielectric CMP, a multilayer Nb–M–Nb–M–Nb–M–Nb will be deposited on the planar surface and patterned to form a vertical stack of three JJs, LJ, where M is the barrier metal (c). Finally all LJ stacks will be planarized to their tops, similar to the JJ planari- zation described in the text. Top wiring will be deposited and patterned to interconnect the JJ-inductors, not shown here. Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 481 Sergey K. Tolpygo a primary goal. However, it cannot be scaled to VLSI until serial-biasing and current-recycling VLSI technology are developed. The main scalability problem of SFQ digital electronics stems from its advantages — magnetic flux information encoding and SFQ voltage pulse transferring, resulting in large-area SFQ cells. It is clear that confining magnetic flux takes much larger volume and effort than confining charge in the gates of CMOS transistors. As a result, the complexity of SCE digital electronics is ex- pected to remain relatively low unless unforeseen break- throughs happen. Superconductor electronics has many advantages in ap- plications where the cryogenic environment is mandatory and dictated by the system performance requirements that cannot be met by any other means. For instance: as control electronics and cryogenic data processors for very large arrays of cryogenic sensors; control electronics for analog quantum computing based on superconducting quantum annealers; for gate-based quantum computing with super- conducting qubits; and for application-specific ultrafast circuits. There is no doubt that (nearly) reversible super- conducting circuits, approaching and crossing the thermo- dynamic limit, will soon be demonstrated as well as many other interesting circuits. However, the impact of SFQ electronics on general-purpose and high-performance computing, in my opinion, will remain low in the foreseea- ble future because of the insufficient scale of integration. Acknowledgment I am very grateful to all my colleagues at MIT Lincoln Laboratory who are involved with fabrication process de- velopment for superconductor electronics, especially to Vladimir Bolkhovsky and Scott Zarr, to Terry Weir and Alex Wynn for their part in device testing, and to Mark Gouker and Leonard Johnson for the discussions and man- agement of the program. I would like to thank Vasili K. Semenov, Alex F. Kirichenko, Timur Filippov, Quentin Herr, Marc Manheimer, and D. Scott Holmes for useful discussions. My special thanks are to Daniel E. Oates for reading the manuscript and suggesting numerous im- provements. This research is based upon work supported by the Of- fice of the Director of National Intelligence (ODNI), Intel- ligence Advanced Research Projects Activity (IARPA), via Air Force Contract FA872105C0002. The views and con- clusions contained herein are those of the author and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or im- plied, of the ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon. 1. Kirill Borisovich Tolpygo, K stoletiyu so dnya rozhdeniya, Fiz. Tv. Tela 58, 1036 (2016). 2. V.V. Rumyanysev and K.V. Gumennyk, J. Photonic Mater. Technol. 1, 1 (2015). 3. A.A. Borgardt, Fizika i Tehnika Vysokih Davleniy 11, No. 4, 19 (2001); A.A. Borgardt and V.A. Telezhkin, ibid. 6, No. 3, 7 (1996). 4. Kirill Borisovich Tolpygo (On his 75th Birthday), Ukr. Fiz. Z. 36, 1279 (1991). 5. A.A. Galkin, Z.S. Gribnikov, S.I. Pekar, E.I. Rashba, O.V. Snitko, and V.I. Sheka, Fiz. Tv. Tela 10, No. 6 (1976). 6. K.B. Tolpygo, Studia Biophysica 69, 35 (1978). 7. K.B. Tolpygo, Biofizika 27, 95 (1982). 8. V.A. Telezhkin, K.B. Tolpygo, and S.K. Tolpygo, Ukr. Fiz. Z. 24, 1679 (1979). 9. S.V. Bespalova, V.A. Telezhkin, K.B. Tolpygo, and S.K. Tolpygo, Ukr. Fiz. Z. 25, 1858 (1980). 10. K.B. Tolpygo, J. Mol. Struct. 299, 185 (1993). 11. S.V. Bespalova and K.B. Tolpygo, J. Mol. Struct. 291, 245 (1991). 12. S.V. Bespalova and K.B. Tolpygo, J. Math. Chem. 18, 179 (1995). 13. R.F. Service, Science 335, 394 (2012). 14. Top500. www.top500.org 15. The Green500 List. www.green500.org 16. D.S. Holmes, A.L. Ripple, and M.A. Manheimer, IEEE Trans. Appl. Supercond. 23, No. 3, 1701610 (2013). 17. F. Bedard, N. Welker, G.R. Gotter, M.A. Escavage, and J.T. Pinkston, Superconducting Technology Assessment, National Security Agency Office Corp. Assessments, Fort Meade, MD, USA (2005). 18. M.A. Manheimer, IEEE Trans. Appl. Supercond. 25, No. 3, 1301704 (2015). 19. L.S. Yu, C.J. Berry, R.E. Drake, K. Li, R.M. Patt, M. Radparvar, S.R. Whitley, and S.M. Faris, IEEE Trans. Magn. 23, 1476 (1987). 20. D. Yohannes, S. Sarwana, S.K. Tolpygo, A. Sahu, Yu.A. Polyakov, and V.K. Semenov, IEEE Trans. Appl. Supercond. 15, No. 2, 90 (2005). 21. D. Yohannes, A. Kirichenko, S. Sarwana, and S.K. Tolpygo, IEEE Trans. Appl. Supercond. 17, No. 2, 181 (2007). 22. S.K. Tolpygo, D. Yohannes, R.T. Hunt, J.A. Vivalda, D. Donnelly, D. Amparo, A.F. Kirichenko, IEEE Trans. Appl. Supercond. 17, No. 2, 946 (2007). 23. S. Nagasawa, Y. Hashimota, H. Numata, and S. Tahara, IEEE Trans. Appl. Supercond. 5, No. 2, 2447 (1995). 24. H. Numata and S. Tahara, IEICE Trans. Electron. E84-C, 2 (2001). 25. S. Nagasawa, K. Hinode, T. Satoh, H. Akaike, Y. Kitagawa, and M. Hidaka, Physica C 412–414, 1429 (2004). 26. S. Nagasawa, T. Satoh, K. Hinode, Y. Kitagawa, M. Hidaka, H. Akaike, A. Fujimaki, K. Takagi, N. Takagi, N. Yoshikawa, Physica C 469, 1578 (2009). 27. S. Nagasawa, K. Hinode, T. Satoh, M. Hidaka, H. Akaike, A. Fujimaki, N. Yoshikawa, S. Nagasawa, K. Takagi, and N. Takagi, IEICE Trans. Electron. E97-C, No. 3, 132 (2014). 482 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 http://dx.doi.org/10.1016/0022-2860(93)80293-5 http://dx.doi.org/10.1016/0022-2860(93)85048-Y http://dx.doi.org/10.1007/BF01164657 http://www.top500.org/ http://www.green500.org/ http://dx.doi.org/10.1109/TASC.2013.2244634 http://dx.doi.org/10.1109/TASC.2013.2244634 http://www.dtic.mil/dtic/tr/fulltext/u2/a464659.pdf http://dx.doi.org/10.1109/TASC.2015.2399866 http://dx.doi.org/10.1109/TASC.2005.849701 http://dx.doi.org/10.1109/TASC.2005.849701 http://dx.doi.org/10.1109/TASC.2007.897399 http://dx.doi.org/10.1109/TASC.2007.897399 http://dx.doi.org/10.1109/TASC.2007.898571 http://dx.doi.org/10.1109/TASC.2007.898571 http://dx.doi.org/10.1109/TASC.2007.898571 http://dx.doi.org/10.1109/77.403086 http://dx.doi.org/10.1016/j.physc.2003.12.097 http://dx.doi.org/10.1016/j.physc.2009.05.219 http://dx.doi.org/10.1587/transele.E97.C.132 Superconductor digital electronics: scalability and energy efficiency issues 28. S.K. Tolpygo, V. Bolkhovsky, T.J. Weir, L.M. Johnson, M.A. Gouker, and W.D. Oliver, IEEE Trans. Appl. Supercond. 25, No. 3, 1101312 (2015). 29. S.K. Tolpygo, V. Bolkhovsky, T.J. Weir, CJ Galbraith, L.M. Johnson, M. Gouker, and V.K. Semenov, IEEE Trans. Appl. Supercond. 25, No. 3, 1100905 (2015). 30. S.K. Tolpygo, V. Bolkhovsky, T.J. Weir, A. Wynn, D.E. Oates, L.M. Johnson, and M.A. Gouker, IEEE Trans. Appl. Supercond. 26, No. 3 (2016). 31. M.W. Johnson, P. Bunyk, F. Maibaum, E. Tolkacheva, A.J. Berkley, E.M. Chapple, R. Harris, J. Johansson, T. Lanting, I. Perminov, E. Ladizinsky, T. Oh, and G. Rose, Supercond. Sci. Technol. 23, No. 6, 065004 (2010). 32. R. Harris, J. Johnson, A.J. Berkley, M.W. Johnson, T. Lanting, S. Han, P. Bunyk, E. Ladizinsky, T. Oh, I. Perminov, E. Tolkacheva, S. Uchaikin, E.M. Chapple, C. Enderud, C. Rich, M. Thom, J. Wang, B. Wilson, and G. Rose, Phys. Rev. B 81, 134510 (2010). 33. P.I. Bunyk, E.M. Hoskinson, M.W. Johnson, E. Tolkacheva, F. Altomare, A.J. Berkley, R. Harris, J.P. Hilton, T. Lanting, A.J. Przybysz, and J. Whittaker, IEEE Trans. Appl. Supercond. 24, No. 4, 1700110 (2014). 34. T. Lanting, A.J. Przybysz, A.Yu. Smirnov, F. M. Spedalieri, M.H. Amin, A. J. Berkley, R. Harris, F. Altomare, S. Boixo, P. Bunyk, N. Dickson, C. Enderud, J. P. Hilton, E. Hoskinson, M. W. Johnson, E. Ladizinsky, N. Ladizinsky, R. Neufeld, T. Oh, I. Perminov, C. Rich, M. C. Thom, E. Tolkacheva, S. Uchaikin, A. B. Wilson, and G. Rose, Phys. Rev. X 4, 021041 (2014). 35. O.A. Mukhanov, IEEE Trans. Appl. Supercond. 3, No. 1, 2578 (1993). 36. O.A. Mukhanov, IEEE Trans. Appl. Supercond. 3, No. 4, 3102 (1993). 37. O.A. Mukhanov and A.F. Kirichenko, IEEE Trans. Appl. Supercond. 5, No. 2, 2461 (1995). 38. S.V. Rylov and R.P. Robertazzi, IEEE Trans. Appl. Supercond. 5, No. 2, 2260 (1995). 39. A.F. Kirichenko, V.K. Semenov, Y.K. Kwong, and V. Nadakumar, IEEE Trans. Appl. Supercond. 5, No. 2, 2857 (1995). 40. S.B. Kaplan, A.F. Kirichenko, O.A. Mukhanov, and S. Sarwana, IEEE Trans. Appl. Supercond. 11, No. 1, 513 (2001). 41. O.A. Mukhanov, V.K. Semenov, W. Li, T.V. Filippov, D. Gupta, A.M. Kadin, D.K. Brock, A.K. Kirichenko, Y.A. Polyakov, and I.V. Vernik, IEEE Trans. Appl. Supercond. 11, No. 1, 601 (2001). 42. A. Kirichenko, S. Sarwana, D. Gupta, I. Rochwarger, and O. Mukhanov, IEEE Trans. Appl. Supercond. 13, No. 2, 454 (2003). 43. D. Gupta, T.V. Filippov, A.F. Kirichenko, D.E. Kirichenko, I.V Vernik, A. Sahu, S. Sarwana, P. Shevchenko, A. Talalaevskii, and O.A. Mukhanov, IEEE Trans. Appl. Supercond. 17, No. 2, 430 (2007). 44. I.V. Vernik, D.E. Kirichenko, T.V. Filippov, A. Talalaevskii, A. Sahu, A. Inamdar, A.F. Kirichenko, D. Gupta, and O.A. Mukhanov, IEEE Trans. Appl. Supercond. 17, No. 2, 442 (2007). 45. A. Sahu, T.V. Filippov, and D. Gupta, IEEE Trans. Appl. Supercond. 19, No. 3, 585 (2009). 46. A. Inamdar, S. Rylov, A. Talalaevskii, A. Sahu, S. Sarwana, D.E. Kirichenko, I.V. Vernik, T.V. Filippov, and D. Gupta, IEEE Trans. Appl. Supercond. 19, No. 3, 670 (2009). 47. T.V. Filippov, M. Dorojevets, A. Sahu, A. Kirichenko, C. Ayala, and O. Mukhanov, IEEE Trans. Appl. Supercond. 21, No. 3, 847 (2011). 48. A.F. Kirichenko, I.V. Vernik, J.A. Vivalda, R.T. Hunt, and D.T. Yohannes, IEEE Trans. Appl. Supercond. 25, No. 3, 1300505 (2015). 49. S. Nagasawa, Y. Hashimoto, H. Numata, and S. Tahara, IEEE Trans. Appl. Supercond. 5, No. 2, 2447 (1995). 50. M. Ito, K. Kawasaki, N. Yoshikawa, A. Fujimaki, H. Terai, and S. Yorozu, IEEE Trans. Appl. Supercond. 15, No. 2, 255 (2005). 51. T. Yamada, M. Yoshida, T. Hanai, A. Fujimaki, H. Hayakawa, Y. Kameda, S. Yorozu, H. Terai, and N. Yoshikawa, IEEE Trans. Appl. Supercond. 15, No. 2, 324 (2005). 52. Y. Hashimoto, S. Yorozu, Y. Kameda, A. Fujimaki, H. Terai, and N. Yoshikawa, IEEE Trans. Appl. Supercond. 15, No. 2, 356 (2005). 53. M. Tanaka, T. Kondo, N. Nakajima, T. Kawamoto, Y. Kamiya, A. Fujimaki, IEEE Trans. Appl. Supercond. 15, No. 2, 400 (2005). 54. Y. Kameda, S. Yorozu, Y Hashimoto, H. Terai, A. Fujimaki, and N. Yoshikawa, IEEE Trans. Appl. Supercond. 15, No. 2, 423 (2005). 55. K. Fujiwara, N. Nakajima, T. Nishigai, M. Ito, N. Yoshikawa, A. Fujimaki, H. Terai, and S. Yorozu, IEEE Trans. Appl. Supercond. 15, No. 2, 427 (2005). 56. K. Saitoh and F. Furuta, IEEE Trans. Appl. Supercond. 15, No. 2, 445 (2005). 57. Y. Yamanashi, M. Tanaka, A. Akimoto, H. Park, Y. Kamiya, N. Irie, N. Yoshikawa, A. Fujimaki, H. Terai, and Y. Hashimoto, IEEE Trans. Appl. Supercond. 17, No. 2, 474 (2007). 58. H. Park, Y. Yamanashi, K. Taketomi, N. Yoshikawa, M. Tanaka, K. Obata, Y. Ito, A. Fujimaki, N. Takagi, K. Takagi, and S. Nagasawa, IEEE Trans. Appl. Supercond. 19, No. 3, 634 (2009). 59. H. Hara, K. Obata, H. Park, Y. Yamanashi, K. Taketomi, N. Yoshikawa, M. Tanaka, A. Fujimaki, N. Takagi, K. Takagi, and S. Nagasawa, IEEE Trans. Appl. Supercond. 19, No. 3, 657 (2009). 60. I. Kataeva, H. Akaike, A. Fujimaki, N. Yoshikawa, N. Takagi, K. Inoue, H. Honda, and K. Murakami, IEEE Trans. Appl. Supercond. 19, No. 3, 665 (2009). 61. I. Kataeva, H. Akaike, A. Fujimaki, N. Yoshikawa, S. Nagasawa, and N. Takagi, IEEE Trans. Appl. Supercond. 19, No. 3, 809 (2009). 62. F. Miyaoka, T. Kainuma, Y. Shimamura, Y. Yamanashi, and N. Yoshikawa, IEEE Trans. Appl. Supercond. 21, No. 3, 823 (2011). Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 483 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Johnson,%20L.M..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Johnson,%20L.M..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Johnson,%20L.M..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Oliver,%20W.D..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Oliver,%20W.D..QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2014.2369213 http://dx.doi.org/10.1109/TASC.2014.2369213 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Wynn,%20A..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Wynn,%20A..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Wynn,%20A..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Johnson,%20L.M..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Johnson,%20L.M..QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2016.2519388 http://dx.doi.org/10.1109/TASC.2016.2519388 http://arxiv.org/find/quant-ph/1/au:+Tolkacheva_E/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Berkley_A/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Berkley_A/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Chapple_E/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Harris_R/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Johansson_J/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Lanting_T/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Perminov_I/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Ladizinsky_E/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Oh_T/0/1/0/all/0/1 http://arxiv.org/find/quant-ph/1/au:+Rose_G/0/1/0/all/0/1 http://dx.doi.org/10.1088/0953-2048/23/6/065004 http://dx.doi.org/10.1088/0953-2048/23/6/065004 http://dx.doi.org/10.1103/PhysRevB.81.134510 http://dx.doi.org/10.1109/TASC.2014.2318294 http://dx.doi.org/10.1109/TASC.2014.2318294 https://journals.aps.org/search/field/author/F.%E2%80%89M.%20Spedalieri https://journals.aps.org/search/field/author/M.%E2%80%89H.%20Amin https://journals.aps.org/search/field/author/A.%E2%80%89J.%20Berkley https://journals.aps.org/search/field/author/R.%20Harris https://journals.aps.org/search/field/author/F.%20Altomare https://journals.aps.org/search/field/author/S.%20Boixo https://journals.aps.org/search/field/author/P.%20Bunyk https://journals.aps.org/search/field/author/N.%20Dickson https://journals.aps.org/search/field/author/C.%20Enderud https://journals.aps.org/search/field/author/J.%E2%80%89P.%20Hilton https://journals.aps.org/search/field/author/E.%20Hoskinson https://journals.aps.org/search/field/author/M.%E2%80%89W.%20Johnson https://journals.aps.org/search/field/author/E.%20Ladizinsky https://journals.aps.org/search/field/author/N.%20Ladizinsky https://journals.aps.org/search/field/author/R.%20Neufeld https://journals.aps.org/search/field/author/T.%20Oh https://journals.aps.org/search/field/author/T.%20Oh https://journals.aps.org/search/field/author/I.%20Perminov https://journals.aps.org/search/field/author/C.%20Rich https://journals.aps.org/search/field/author/M.%E2%80%89C.%20Thom https://journals.aps.org/search/field/author/E.%20Tolkacheva https://journals.aps.org/search/field/author/S.%20Uchaikin https://journals.aps.org/search/field/author/S.%20Uchaikin https://journals.aps.org/search/field/author/A.%E2%80%89B.%20Wilson https://journals.aps.org/search/field/author/G.%20Rose http://dx.doi.org/10.1103/PhysRevX.4.021041 http://dx.doi.org/10.1109/77.233529 http://dx.doi.org/10.1109/77.251810 http://dx.doi.org/10.1109/77.403089 http://dx.doi.org/10.1109/77.403089 http://dx.doi.org/10.1109/77.403035 http://dx.doi.org/10.1109/77.403035 http://dx.doi.org/10.1109/77.403187 http://dx.doi.org/10.1109/77.919395 http://dx.doi.org/10.1109/TASC.2003.813898 http://dx.doi.org/10.1109/TASC.2007.898255 http://dx.doi.org/10.1109/TASC.2007.898255 http://dx.doi.org/10.1109/TASC.2007.898613 http://dx.doi.org/10.1109/TASC.2009.2018061 http://dx.doi.org/10.1109/TASC.2009.2018061 http://dx.doi.org/10.1109/TASC.2009.2017767 http://dx.doi.org/10.1109/TASC.2010.2103918 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Hunt,%20R.T..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yohannes,%20D.T..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yohannes,%20D.T..QT.&newsearch=true http://dx.doi.org/10.1109/77.403086 http://dx.doi.org/10.1109/TASC.2005.849773 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Fujimaki,%20A..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Fujimaki,%20A..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kameda,%20Y..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kameda,%20Y..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Terai,%20H..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yoshikawa,%20N..QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yoshikawa,%20N..QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2005.849835 http://dx.doi.org/10.1109/TASC.2005.849860 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.H.%20Terai.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.A.%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Yoshikawa.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2005.849865 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.M.%20Ito.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.A.%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.H.%20Terai.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.S.%20Yorozu.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2005.849866 http://dx.doi.org/10.1109/TASC.2005.849866 http://dx.doi.org/10.1109/TASC.2005.849870 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.H.%20Park.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20Kamiya.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Irie.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Irie.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.A.%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.H.%20Terai.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20Hashimoto.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2007.898606 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Masamitsu%20Tanaka.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Masamitsu%20Tanaka.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Koji%20Obata.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yuki%20Ito.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Akira%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Naofumi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kazuyoshi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Shuichi%20Nagasawa.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2009.2019070 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yuki%20Yamanashi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kazuhiro%20Taketomi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Masamitsu%20Tanaka.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Akira%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kazuyoshi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.S.%20Nagasawa.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2009.2018039 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Naofumi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Naofumi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Koji%20Inoue.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Hiroaki%20Honda.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kazuaki%20Murakami.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2009.2018039 http://dx.doi.org/10.1109/TASC.2009.2018039 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.S.%20Nagasawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Takagi.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2010.2092402 http://dx.doi.org/10.1109/TASC.2010.2092402 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yuki%20Yamanashi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2010.2094175 Sergey K. Tolpygo 63. T. Kainuma, Y. Shimamura, F. Miyaoka, Y. Yamanashi, N. Yoshikawa, A. Fujimaki, K. Takagi, N. Takagi, and S. Nagasawa, IEEE Trans. Appl. Supercond. 21, No. 3, 827 (2011). 64. R. Nakamoto, S. Sakuraba, T. Sato, and K. Nakajima, IEEE Trans. Appl. Supercond. 21, No. 3, 852 (2011). 65. M. Dorojevets, C.L. Ayala, N. Yoshikawa, A. Fujimaki, IEEE Trans. Appl. Supercond. 23, No. 3, June 2013, Art. ID 1700605. 66. M. Dorojevets, A.K. Kasperek, N. Yoshikawa, and A. Fujimaki, IEEE Trans. Appl. Supercond. 23, No. 3, 1300104 (2013). 67. M. Dorojevets, C.L. Ayala, N. Yoshikawa, and A. Fujimaki, IEEE Trans. Appl. Supercond. 23, No. 3, 1700104 (2013). 68. M. Hidaka, S. Nagasawa, K. Hinode, and T. Satoh, IEEE Trans. Appl. Supercond. 23, No. 3, 1100906 (2013). 69. X. Peng, Q. Xu, T. Kato, Y. Yamanashi, N. Yoshikawa, A. Fujimaki, N. Takagi, K. Takagi, and M. Hidaka, IEEE Trans. Appl. Supercond. 25, No. 3, 1301106 (2015). 70. Y. Sakashita, Y. Yamanashi, and N. Yoshikawa, IEEE Trans. Appl. Supercond. 25, No. 3, 1301205 (2015). 71. X. Peng, Y. Shimamura, Y. Yamanashi, N. Yoshikawa, A. Fujimaki, K. Takagi, N. Takagi, and M. Hidaka, Proceedings of 14th Intern. Supercond. Electron. Conf., ISEC’2013, p. 35, July 7–11, 2013, Cambridge, Massachusetts. 72. 72. T. Kato, Y. Yamanashi, N. Yoshikawa, A. Fujimaki, N. Takagi, K. Takagi, and S. Nagasawa, Proceedings of 14th Intern. Supercond. Electron. Conf., ISEC’2013, p. 56, Cambridge, Massachusetts, July 7–11, 2013. 73. A. Akitomo, Y. Yamanashi, and N. Yoshikawa, Proceedings of 14th Intern. Supercond. Electron. Conf., ISEC’2013, p. 102, Cambridge, Massachusetts, July 7–11, 2013. 74. M. Tanaka, K. Takata, R. Sato, A. Fujimaki, T. Kawaguchi, Y. Ando, K. Takagi, N. Takagi, and N. Yoshikawa, Proceedings of 15th Intern. Supercond. Electron. Conf., ISEC’2015, p. DS-001-INV, July 6–9, 2015, Nagoya, Japan. 75. V.K. Semenov, Y.A. Polyakov, and S.K. Tolpygo, IEEE Trans. Appl. Supercond. 25, 1301507 (2015). 76. Q.P. Herr, J. Osborne, M.J.A. Stoutimore, H. Hearne, R. Selig, J. Vogel, E. Min, V.V. Talanov, and A.Y. Herr, Supercond. Sci. Technol. 28, 124003 (2015). 77. P. Bunyk, K. Likharev, and D. Zinoviev, Int. J. High Speed Electron. Syst. 11, 257 (2001). 78. K.K. Likharev, O.A. Mukhanov, and V.K. Semenov, IEEE Trans. Magn. 23, 759 (1987). 79. K.K. Likharev and V.K. Semenov, IEEE Trans. Appl. Supercond. 1, No. 1, 3 (1991). 80. D.F. Kirichenko, S. Sarwana, and A.F. Kirichenko, IEEE Trans. Appl. Supercond. 21, No. 3, 776 (2011). 81. M.H. Volkmann, A. Sahu, C.J. Fourie, and O.A. Mukhanov, Supercond. Sci. Technol. 26, No. 1, 015002 (2013). 82. Q.P. Herr, A.Y. Herr, O.T. Oberg, and A.G. Ioannidis, J. Appl. Phys. 109, 103903 (2011). 83. M. Hosoya, W. Hioe, and J. Casas, R. Kamikawai, Y. Harada, Y. Wada, H. Nakane, R. Suda, and E. Goto, IEEE Trans. Appl. Supercond. 1, No. 2, 77 (1991). 84. Y. Harada, H. Nakane, N. Miyamoto, U. Kawabe, E. Goto, and T. Soma, IEEE Trans. Magn. MAG-23, 3801 (1987). 85. N. Takeuchi, D. Ozawa, Y. Tamanashi, and N. Yoshikawa, Supercond. Sci. Technol. 26, No. 3, 035010 (2013). 86. K. Inoue, N. Takeuchi, K. Ehara, Y. Yamanashi, and N. Yoshikawa, IEEE Trans. Appl. Supercond. 23, No. 3, 1301105 (2013). 87. L.A. Abelson and G.L. Kerber, Proc. IEEE 92, 1517 (2004). 88. M. Gurvitch, M.A. Washingoton, and H.A. Huggins, Appl. Phys. Lett. 42, 472 (1983). 89. D.E. McCumber, J. Appl. Phys. 39, 3113 (1968). 90. W.C. Stewart, Appl. Phys. Lett. 12, 277 (1968). 91. RSFQ @ SUNY Stony Brook. 92. V. Patel and J.E. Lukens, IEEE Trans. Appl. Supercond. 9, No. 2, 3247 (1999). 93. A.V. Rylyakov and K.K. Likharev, IEEE Trans. Appl. Supercond. 9, No. 2, 3539 (1999). 94. V.K. Semenov and M.A. Voronova, IEEE Trans. Magn. 23, 1476 (1987). 95. V. Semenov and Y. Polyakov, IEEE Trans. Appl. Supercond. 11, No. 1, 550 (2001). 96. J.H. Kang and S.B. Kaplan, IEEE Trans. Appl. Supercond. 13, No. 2, 547 (2003). 97. T.V. Filippov, A. Sahu, S. Sarwana, D. Gupta, and V.K. Semenov, IEEE Trans. Appl. Supercond. 19, No. 3, 580 (2009). 98. M. Igarashi, Y. Yamanashi, N. Yoshikawa, Kan Fujiwara, and Yoshihito Hashimoto, IEEE Trans. Appl. Supercond. 19, No. 3, 649 (2009). 99. V.K. Semenov, Y.A. Polyakov, and S.K. Tolpygo, Submitted to ASC (2016), unpublished. 100. B.D. Josephson, Phys. Lett. 1, 251 (1962). 101. R. Landauer, IMB J. Res. Dev. 5, 183 (1961). 102. List of CPU power dissipation figures. 103. Transistor count. 104. M. Dorojevets, Z. Chen, C.L. Ayala, and A.K. Kasperek, IEEE Trans. Appl. Supercond. 25, No. 3, 1300408 (2015). 105. V.K. Semenov, IEEE Trans. Appl. Supercond. 23, No. 3, 1700908 (2011). 106. W. Chen, A.V. Rylyakov, V. Patel, J.E. Lukens, and K.K. Likharev, Appl. Phys. Lett. 73, 2817 (1998). 107. V.K. Semenov, G.V. Danilov, and D.V. Averin, IEEE Trans. Appl. Supercond. 13, No. 2, 938 (2003). 108. V.K. Semenov, G.V. Danilov, and D.V. Averin, IEEE Trans. Appl. Supercond. 17, No. 2, 455 (2007). 109. J. Ren, V.K. Semenov, Y.A. Polyakov, V. Averin, and J.-S. Tsai, IEEE Trans. Appl. Supercond. 19, No. 3, 961 (2009). 110. J. Ren and V.K. Semenov, IEEE Trans. Appl. Supercond. 21, No. 3, 780 (2011). 111. K. Senapati, M.G. Blamire, and Z.H. Barber, Nat. Mater. 10, 849 (2011). 112. V. Ryazanov, V.V. Bol’ginov, D.S. Sobanin, I.V. Vernik, S.K. Tolpygo, A.M. Kadin, and O.A. Mukhanov, Phys. Proc. 36, 35 (2012). 484 Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yuki%20Yamanashi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Akira%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kazuyoshi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Naofumi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Shuichi%20Nagasawa.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2010.2096374 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Koji%20Nakajima.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2010.2095814 http://dx.doi.org/10.1109/TASC.2010.2095814 http://dx.doi.org/10.1109/TASC.2012.2233846 http://dx.doi.org/10.1109/TASC.2012.2227648 http://dx.doi.org/10.1109/TASC.2012.2229334 http://dx.doi.org/10.1109/TASC.2012.2237471 http://dx.doi.org/10.1109/TASC.2012.2237471 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yuki%20Yamanashi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Nobuyuki%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Akira%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Akira%20Fujimaki.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Naofumi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kazuyoshi%20Takagi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Mutsuo%20Hidaka.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2014.2382973 http://dx.doi.org/10.1109/TASC.2014.2382973 http://dx.doi.org/10.1109/TASC.2014.2384833 http://dx.doi.org/10.1109/TASC.2014.2384833 http://dx.doi.org/10.1109/ISEC.2015.7383449 http://dx.doi.org/10.1109/ISEC.2015.7383449 http://dx.doi.org/10.1109/TASC.2014.2382665 http://dx.doi.org/10.1109/TASC.2014.2382665 http://dx.doi.org/10.1088/0953-2048/28/12/124003 http://dx.doi.org/10.1142/S012915640100085X http://dx.doi.org/10.1142/S012915640100085X http://dx.doi.org/10.1109/77.80745 http://dx.doi.org/10.1109/77.80745 http://dx.doi.org/10.1109/TASC.2010.2098432 http://dx.doi.org/10.1109/TASC.2010.2098432 http://dx.doi.org/10.1088/0953-2048/26/1/015002 http://dx.doi.org/10.1063/1.3585849 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.R.%20Kamikawai.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20Harada.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20Wada.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.H.%20Nakane.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.R.%20Suda.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.E.%20Goto.QT.&newsearch=true http://dx.doi.org/10.1109/77.84613 http://dx.doi.org/10.1109/77.84613 http://dx.doi.org/10.1088/0953-2048/26/3/035010 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20Yamanashi.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Yoshikawa.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.N.%20Yoshikawa.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2012.2236133 http://dx.doi.org/10.1109/JPROC.2004.833652 http://dx.doi.org/10.1063/1.93974 http://dx.doi.org/10.1063/1.93974 http://dx.doi.org/10.1063/1.1656743 http://dx.doi.org/10.1063/1.1651991 http://www.physics.sunysb.edu/Physics/RSFQ/index.html http://dx.doi.org/10.1109/77.783721 http://dx.doi.org/10.1109/77.783794 http://dx.doi.org/10.1109/77.783794 http://dx.doi.org/10.1109/77.919404 http://dx.doi.org/10.1109/77.919404 http://dx.doi.org/10.1109/TASC.2003.813932 http://dx.doi.org/10.1109/TASC.2003.813932 http://dx.doi.org/10.1109/TASC.2009.2018426 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Kan%20Fujiwara.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Yoshihito%20Hashimoto.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2009.2018084 http://dx.doi.org/10.1016/0031-9163(62)91369-0 https://en.wikipedia.org/wiki/List_of_CPU_power_dissipation_figures https://en.wikipedia.org/wiki/Transistor_count http://dx.doi.org/10.1109/TASC.2014.2368354 http://dx.doi.org/10.1109/TASC.2012.2237214 http://dx.doi.org/10.1063/1.122600 http://dx.doi.org/10.1109/TASC.2003.814155 http://dx.doi.org/10.1109/TASC.2003.814155 http://dx.doi.org/10.1109/TASC.2007.898260 http://dx.doi.org/10.1109/TASC.2007.898260 http://dx.doi.org/10.1109/tasc.2009.2018250 http://dx.doi.org/10.1109/TASC.2011.2104352 http://dx.doi.org/10.1038/nmat3116 http://dx.doi.org/10.1016/j.phpro.2012.06.126 http://dx.doi.org/10.1016/j.phpro.2012.06.126 Superconductor digital electronics: scalability and energy efficiency issues 113. T.A. Larkin, V.V. Bol’ginov, V.S. Stolyarov, V.V. Ryazanov, I.V. Vernik,. S.K. Tolpygo, and O.A. Mukhanov, Appl. Phys. Lett. 100, 222601 (2012). 114. S. Nagasawa, K. Hinode, T. Satoh, Y. Kitagawa, and M. Hidaka, Supercond. Sci. Technol. 19, S325 (2006). 115. S. Nagasawa, T. Satoh, Y. Kitagawa, and M. Hidaka, IEEE Trans. Appl. Supercond. 17, No. 2, 177 (2007). 116. S.K. Tolpygo, D. Tolpygo, R.T. Hunt, S. Narayana, Y.A. Polyakov, and V.K. Semenov, IEEE Trans. Appl. Supercond. 19, No. 3, 598 (2009). 117. S. Narayana, V.K. Semenov, Y.A. Polyakov, V. Dotsenko, and S.K. Tolpygo, Supercond. Sci. Technol. 25, 105012 (2012). 118. O. Mukhanov, IEEE Trans. Appl. Supercond. 21, No. 3, 760 (2011). 119. D.K. Brock, A.M. Kadin, A.F. Kirichenko, O.A. Mukhanov, S. Sarwana, J.A. Vivalda, W. Chen, and J.E. Lukens, IEEE Trans. Appl. Supercond. 11, No. 2, 369 (2001). 120. R.E. Miller, W.H. Mallison, A.W. Kleinsasser, K.A. Delin, and E.M. Macedo, Appl. Phys. Lett. 63, 1423 (1993). 121. A.W. Kleinsasser, R.E. Miller, W.H. Mallison, and G.B. Arnold, Phys. Rev. Lett. 72, 1738 (1994). 122. A.W. Kleinsasser, IEEE Trans. Appl. Supercond. 11, No. 1, 1043 (2001). 123. S.K. Tolpygo, D.J.C. Amparo, R.T. Hunt, J.A. Vivalda, and D.T. Yohannes, IEEE Trans. Appl. Supercond. 23, No. 3, 1100305 (2013). 124. S.K. Tolpygo and D. Amparo, J. Appl. Phys. 104, 063904 (2008). 125. A.W. Kleinsasser and R.A. Burhman, Appl. Phys. Lett. 37, 841 (1980). 126. A.W. Kleinsasser, W.H. Mallison, and R.E. Miller, IEEE Trans. Appl. Supercond. 5, No. 2, 2318 (1995). 127. G.L. Kerber, A.W. Kleinsasser, and B. Bumble, IEEE Trans. Appl. Supercond. 19, No. 3, 159 (2009). 128. A.L. Gudkov, M.Yu. Kupriyanov, and A.N. Samus, J. Exp. Theor. Phys. 114, 818 (2012). 129. D. Olaya, B. Baek, P. Dresselhaus, and S. Benz, IEEE Trans. Appl. Supercond. 18, No. 4, 1797 (2008). 130. D. Olaya, P. Dresselhaus, S. Benz, A. Herr, Q.P. Herr, A.G. Ioannidis, D.L. Miller, and A.W. Kleinsasser, Appl. Phys. Lett. 96, 213510 (2010). 131. P. Dresselhaus, M. Elsbury, D. Olaya, C.J. Burroughs, S.P. Benz, IEEE Trans. Appl. Supercond. 21, No. 3, 693 (2011). 132. H. Kroger, C.N. Potter, and D.W. Jillie, IEEE Trans. Magn. 15, 488 (1979). 133. D. Jillie, L.N. Smith, H. Kroger, L.W. Currier, C. Potter, D.M. Shaw, and R.L. Payer, IEEE J. Solid State Circuits 18, 173 (1983). 134. H. Kroger, L. Smith, D. Jillie, J. Thaxter, R. Aucoin, L. Currier, C. Potter, D. Shaw, and P. Willis, IEEE Trans. Magn. 21, 870 (1985). 135. I.A. Devyatov and M.Yu. Kuprianov, JETP Lett. 59, 200 (1994). 136. Y. Naveh, V. Patel, D.V. Averin, K.K. Likharev, and J.E. Lukens, Phys. Rev. Lett. 85, 5404 (2000). 137. V. Patel, Current Transport Properties of Nb/AlOx/Nb Josephson Junctions with High Barrier Transparencies, PhD Thesis, State University of New York at Stony Brook, NY (2001). 138. S.K. Tolpygo, V. Bolkhovsky, T. Weir, L. Johnson, W.D. Oliver, and M.A. Gouker, Supercond. Sci. Technol. 27, 025016 (2014). 139. J.E. Mooij, T.P. Orlando, L. Levitov, L. Tian, C.H. van der Wal, and S. Lloyd, Science 285, 1036 (1999). 140. T.P. Orlando, J.E. Mooij, L. Tian, C.H. Van der Wal, L. Levitov, S. Lloyd, and J.J. Mazo, Phys. Rev. B 60, 15398 (1999). 141. V.E. Manucharyan, J. Koch, L.I. Glazman, and M.H. Devoret, Science 326, 113 (2009). 142. V. Kaplunenko and G. Fisher, Supercond. Sci. Technol. 17, S145 (2004). 143. A. Ustinov and V. Kaplunenko, J. Appl. Phys. 94, 5405 (2003). 144. A. Feofanov, V. Oboznov, V. Bol’ginov, J. Lisenfeld, S. Poletto, V.V. Ryazanov, A.N. Rossolenko, M. Khabipov, D. Balashov, A.B. Zorin, P.N. Dmitriev, V.P. Koshelets, and A.V. Ustinov, Nature Phys. 6, 593 (2010). 145. M. Khabibov, D. Balashov, F. Maibaum, M.I. Khabipov, D.V. Balashov, F. Maibaum, A.B. Zorin, V.A. Oboznov, V.V. Bolginov, A.N. Rossolenko, and V.V. Ryazanov, Supercond. Sci. Technol. 23, 045032 (2010). 146. O. Wetzstein, T. Ortlepp, R. Stolz, J. Kunert, H.G. Meyer, and H. Toepfer., IEEE Trans. Appl. Supercond. 21, No. 3, 814 (2011). 147. N. Takeuchi, Y. Yamanashi, and N. Yoshikawa, Appl. Phys. Lett. 102, 052602 (2013). 148. K.K. Likharev, IEEE Trans. Magn. 13, 242 (1977). 149. K.K. Likharev, S.V. Rylov, and V.K. Semenov, IEEE Trans. Magn. 21, 947 (1985). 150. K.K. Likharev, Int. J. Theor. Phys. 21, 311 (1982). 151. Q. Xu, Y. Yamanashi, and N. Yoshikawa, Extended Abstracts of 15th International Superconductive Electronics Conference, ISEC’2015, DS-P21, Nagoya, Japan (2015). 152. H. Namata, S. Nagasawa, and S. Tahara, IEEE Trans. Appl. Supercond. 7, No. 2, 2282 (1997). 153. A.M. Kadin, C.A. Manchini, M.J. Feldman, and D.K. Brock, IEEE Trans. Appl. Supercond. 11, No. 1, 1050 (2001). Low Temperature Physics/Fizika Nizkikh Temperatur, 2016, v. 42, No. 5 485 http://dx.doi.org/10.1063/1.4723576 http://dx.doi.org/10.1063/1.4723576 http://dx.doi.org/10.1088/0953-2048/19/5/S34 http://dx.doi.org/10.1109/TASC.2007.898050 http://dx.doi.org/10.1109/TASC.2007.898050 http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.S.%20Narayana.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20A.%20Polyakov.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Y.%20A.%20Polyakov.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.V.%20K.%20Semenov.QT.&newsearch=true http://dx.doi.org/10.1109/TASC.2009.2017858 http://dx.doi.org/10.1109/TASC.2009.2017858 http://dx.doi.org/10.1088/0953-2048/25/10/105012 http://dx.doi.org/10.1109/TASC.2010.2096792 http://dx.doi.org/http:/dx.doi.org/10.1063/1.109697 http://dx.doi.org/10.1103/PhysRevLett.72.1738 http://dx.doi.org/10.1109/77.919527 http://dx.doi.org/10.1109/TASC.2012.2227438 http://dx.doi.org/10.1063/1.2977725 http://dx.doi.org/10.1063/1.92067 http://dx.doi.org/10.1109/77.403049 http://dx.doi.org/10.1109/77.403049 http://dx.doi.org/10.1109/TASC.2009.2017859 http://dx.doi.org/10.1109/TASC.2009.2017859 http://dx.doi.org/0.1109/TASC.2008.2007652 http://dx.doi.org/0.1109/TASC.2008.2007652 http://dx.doi.org/10.1063/1.3432065 http://dx.doi.org/10.1063/1.3432065 http://dx.doi.org/10.1109/TASC.2010.2079310 http://dx.doi.org/10.1109/TMAG.1985.1063748 http://dx.doi.org/10.1103/PhysRevLett.85.5404 http://dx.doi.org/10.1088/0953-2048/27/2/025016 http://dx.doi.org/10.1126/science.285.5430.1036 http://dx.doi.org/10.1103/PhysRevB.60.15398 http://dx.doi.org/10.1126/science.1175552 http://dx.doi.org/10.1088/0953-2048/17/5/011 http://dx.doi.org/10.1063/1.1604964 http://dx.doi.org/10.1038/nphys1700 http://dx.doi.org/10.1088/0953-2048/23/4/045032 http://dx.doi.org/10.1088/0953-2048/23/4/045032 http://dx.doi.org/10.1109/TASC.2010.2102998 http://dx.doi.org/10.1063/1.4790276 http://dx.doi.org/10.1063/1.4790276 http://dx.doi.org/10.1007/BF01857733 http://dx.doi.org/10.1109/77.621694 http://dx.doi.org/10.1109/77.621694 http://dx.doi.org/10.1109/77.919528 1. Introduction 2. Fabrication and scalability of SFQ circuits 2.1. SFQ electronics fabrication processes 3. Physical constraints on VLSI of SFQ electronics 3.1. Josephson junctions 3.2. Statistical variations of Josephson junctions 3.3. Limitations on scaling caused by bias currents 3.4. Heating of resistors 3.5. Circuit inductors 4. Energy dissipation and efficiency of SCE 4.1. RSFQ-based and RQL circuits 4.2. Adiabatic quantum flux parametron (AQFP) circuits 4.3. Summary 5. Future work 5.1. Energy-efficient computing 5.2. High-speed computing 5.3. VLSI technology development 5.3.1. Self-shunted, high-Jc junctions 5.3.2. Number of junction layers 5.3.3. Compact inductors 5.3.4. JJs as inductors 5.3.5. Phase shifters and pi-junctions 6. Conclusion Acknowledgment