This paper describes our programme to develop and demonstrate ultra-high performance single flux quantum (SFQ) VLSI technology that will enable superconducting digital processors for petaFLOPS-scale computing. In the hybrid technology, multi-threaded architecture, the computational engine to power a petaFLOPS machine at affordable power will consist of 4096 SFQ multi-chip processors, with 50 to 100 GHz clock frequency and associated cryogenic RAM. We present the superconducting technology requirements, progress to date and our plan to meet these requirements. We improved SFQ Nb VLSI by two generations, to a 8 kA cm −2 , 1.25 µm junction process, incorporated new CAD tools into our methodology, demonstrated methods for recycling the bias current and data communication at speeds up to 60 Gb s −1 , both on and between chips through passive transmission lines. FLUX-1 is the most ambitious project implemented in SFQ technology to date, a prototype general-purpose 8 bit microprocessor chip. We are testing the FLUX-1 chip (5K gates, 20 GHz clock) and designing a 32 bit floating-point SFQ multiplier with vector-register memory. We report correct operation of the complete stripline-connected gate library with large bias margins, as well as several larger functional units used in FLUX-1. The next stage will be an SFQ multi-processor machine. Important challenges include further reducing chip supply current and on-chip power dissipation, developing at least 64 kbit, sub-nanosecond cryogenic RAM chips, developing thermally and electrically efficient high data rate cryogenic-to-ambient input/output technology and improving Nb VLSI to increase gate density.
This paper describes our programme to develop and demonstrate ultra-high performance single flux quantum (SFQ) VLSI technology that will enable superconducting digital processors for petaFLOPS-scale computing. In the hybrid technology, multi-threaded architecture, the computational engine to power a petaFLOPS machine at affordable power will consist of 4096 SFQ multi-chip processors, with 50 to 100 GHz clock frequency and associated cryogenic RAM. We present the superconducting technology requirements, progress to date and our plan to meet these requirements. We improved SFQ Nb VLSI by two generations, to a 8 kA cm −2 , 1.25 µm junction process, incorporated new CAD tools into our methodology, demonstrated methods for recycling the bias current and data communication at speeds up to 60 Gb s −1 , both on and between chips through passive transmission lines. FLUX-1 is the most ambitious project implemented in SFQ technology to date, a prototype general-purpose 8 bit microprocessor chip. We are testing the FLUX-1 chip (5K gates, 20 GHz clock) and designing a 32 bit floating-point SFQ multiplier with vector-register memory. We report correct operation of the complete stripline-connected gate library with large bias margins, as well as several larger functional units used in FLUX-1. The next stage will be an SFQ multi-processor machine. Important challenges include further reducing chip supply current and on-chip power dissipation, developing at least 64 kbit, sub-nanosecond cryogenic RAM chips, developing thermally and electrically efficient high data rate cryogenic-to-ambient input/output technology and improving Nb VLSI to increase gate density.
Single flux quantum (SFQ) electronics is extremely fast and has very low on-chip power dissipation. SFQ VLSI is an excellent candidate for high-performance computing and other applications requiring extremely high-speed signal processing. Despite this, SFQ technology has generally not been accepted for system implementation. We argue that this is due, at least in part, to the use of outdated tools to produce SFQ circuits and chips. Assuming the use of tools equivalent to those employed in the semiconductor industry, we estimate the density of Josephson junctions, circuit speed, and power dissipation that could be achieved with SFQ technology. Today, CMOS lithography is at 90?65?nm with about 20 layers. Assuming equivalent technology, aggressively increasing the current density above 100?kA?cm?2 to achieve junction speeds approximately 1000?GHz, and reducing device footprints by converting device profiles from planar to vertical, one could expect to integrate about 250?M Josephson junctions cm?2 into SFQ digital circuits. This should enable circuit operation with clock frequencies above 200?GHz and place approximately 20?K gates within a radius of one clock period. As a result, complete microprocessors, including integrated memory registers, could be fabricated on a single chip.
The Hybrid Technology Multi-Threaded (HTMT) approach to petaflops computing includes large numbers of ultra-high performance Nb Rapid Single Flux Quantum (RSFQ) processor and memory chips, making it by far the largest active superconducting electronics project in the United States. In order to achieve petaflops, RSFQ circuits with 10/sup 5/ to 10/sup 6/ junctions per chip will be required to operate at clock speeds of 50 to 100 GHz, far beyond the current state of the art. In this paper, we review the state of the art of Nb circuit fabrication and discuss the requirements for significantly improving circuit density and speed.
Superconductive electronics and, specifically, single flux quantum (SFQ) circuits are attracting significant attention as a promising ultra-low power and ultra-high speed circuit technology for beyond CMOS exascale supercomputing. Superconductive electronics is suitable for large scale, energy efficient, stationary cloud computing, ”big data” applications, and ultra-high speed neural networks. Significant development in the design and manufacture of superconductive electronics for prospective exascale computing systems has led to device densities exceeding 4.2×10^6 Josephson junctions (JJs) per cm2 – sufficient to enable large scale SFQ systems. The lack of electronic design automation (EDA) tools, however, is currently a serious issue affecting the development of complex SFQ circuits. Advances in EDA algorithms and design and analysis methodologies and techniques are therefore required to enhance the development of VLSI complexity SFQ systems. In this dissertation, issues and solutions to enable, connect, and secure VLSI complexity SFQ system are presented. Topics such as interconnects, ultra-high speed clocking and synchronization, signal interfaces, and deep scaling of superconductive structures are described, and related design methodologies, algorithms, circuits, and expressions are proposed. On-chip signal routing has become an issue of growing importance in modern VLSI complexity SFQ systems. Specialized routing methods for these systems are presented. The routing methods include passive transmission lines as global interconnect and Josephson transmission lines as local interconnect. Guidelines are provided for determining when to use global and local interconnects. Novel methodologies, algorithms, and expressions have been developed to manage resonance, flux trapping, and coupling in VLSI complexity SFQ circuits. The design of an SFQ clock distribution network is a challenging problem due to timing uncertainties at very high frequencies, gate-level pipelining, and fanout constraints. Near zero skew clock tree targeting large scale SFQ circuits is proposed to improve the global timing process while eliminating skew uncertainty. To enable higher fanout, area efficient splitter topologies are developed, producing a clock network suitable for large scale SFQ systems. All-JJ superconductive circuits based on bistable superconductor-ferromagnet-superconductor are presented to enable nanometer feature sizes for VLSI superconductive systems. These specialized guidelines, algorithms, and circuits enable deep scaling of superconductive systems for stationary high speed, ultra-low energy applications.
The first single-chip superconductor FLUX-1 microprocessor has been designed in the Rapid Single Flux Quantum (RSFQ) logic and fabricated using 4 kA/cm 2 , 1.75-μ m Nb/AlOx/Nb Josephson junction technology as a result of the collaboration between SUNY Stony Brook and TRW, Inc. A FLUX-1 chip represents an 8-bit deeply pipelined microprocessor prototype with a target clock frequency of 17-20 GHz. A new parallel partitioned architecture has been developed in order to tolerate interconnect delays and fill long FLUX-1 processor pipelines with useful instructions. The processor includes the 16 × 32-bit pipelined instruction memory, 8 integer arithmetic-logic units interleaved with 8 registers, the branch unit, and I/O ports for 5-GHz chip-to-chip communication over Nb microstrip lines on a chip carrier. The FLUX-1 instruction set consists of ~25 arithmetic, logical, and control instructions. A FLUX-1 microprocessor chip contains 65,759 Josephson junctions on a 10.6 mm × 13.2 mm die with flip-chip packaging. First FLUX-1 chips fabricated in August 2001 are currently under testing at TRW, Inc.
I describe the fundamental technologies concerning the single flux quantum (SFQ) logic circuits. Recent remarkable progress in the computer-aided-design (CAD) tool, which is based on the cell-based design technique, enables us to make SFQ integrated circuits. Although the circuit demonstrated so far has 2, 300 Josephson junctions at most, this number will increase because the CAD tool is advancing. The on-chip test system has also been developed to test the high-speed nature of the SFQ circuit. For example, the shift-register employed as a circuit under test operates correctly at 55GHz. I also describe the potential applications of the SFQ circuits. An analog-to-digital converter utilizing the high-speed nature and high sensitivity in the SFQ circuits will be applicable to the front end of the future base stations, which require broad band width, high sensitivity, and flexibility. Packet switches used in the future high-end router need high throughput and functionality. Only SFQ-based packet switches can meet these requirements. Another potential application is a high-end server. We have studied micro-architectures of a processor executing the Java Byte Codes directly and a multiprocessor system constructed by connecting the processors and memories via the ultra-broad-band SFQ network switches. Throughout the study, we employ the concept of the complexity-reduction (CORE), in which the high-speed nature of the SFQ circuits is used for the reduction of the circuit scale or simplification of the micro-architecture; the multiprocessor system has high performances. These clear applications promote rapid technological progress in the SFQ technology, though we must make every endeavor to find new application fields.
The first single-chip superconductor FLUX-1 microprocessor has been designed in the Rapid Single Flux Quantum (RSFQ) logic and fabricated using 4 kA/cm2, 1.75-μm Nb/AlOx/Nb Josephson junction technology as a result of the collaboration between SUNY Stony Brook and TRW, Inc. A FLUX-1 chip represents an 8-bit deeply pipelined microprocessor prototype with a target clock frequency of 17-20 GHz. A new parallel partitioned architecture has been developed in order to tolerate interconnect delays and fill long FLUX-1 processor pipelines with useful instructions. The processor includes the 16 × 32-bit pipelined instruction memory, 8 integer arithmetic-logic units interleaved with 8 registers, the branch unit, and I/O ports for 5-GHz chip-to-chip communication over Nb microstrip lines on a chip carrier. The FLUX-1 instruction set consists of ~25 arithmetic, logical, and control instructions. A FLUX-1 microprocessor chip contains 65,759 Josephson junctions on a 10.6 mm × 13.2 mm die with flip-chip packaging. First FLUX-1 chips fabricated in August 2001 are currently under testing at TRW, Inc.