Abstract : The major objective of the project was to design and demonstrate operation of key components of a 30 GHz 16-bit RSFQ processor prototype implemented with the AIST/ISTEC 10 kA/cm sq. fabrication process. Our team has developed complete logical and physical designs of five RSFQ chips using the CONNECT cell library and RSFQ CAD tools developed at the Universities of Yokohama and Nagoya (Japan). The major results are the world's first successful design, fabrication, and demonstration of correct operation of a 20 GHz 8x8-bit parallel carry-save RSFQ multiplier with approximately 6K JJs, a 16-bit sparse-tree wave-pipelined RSFQ adder with approximately 10K JJs, and partial operation of an 8-bit ALU chip with approximately 9K JJs. The goal of the second phase of the project was to get detailed understanding of the performance, complexity, and energy efficiency of on-chip storage units implemented with superconductor Reciprocal Quantum Logic (RQL) using our RQL VHDL cell library tuned to the MIT Lincoln Laboratory 10 kA/cm2 248 nm process. The 8.5 GHz 1-4 Kbit 32-/64-bit multi-ported scratchpad memory, register files, write-through and write-back caches designed with RQL Non-Destructive Read-Out storage cells have the average energy consumption of 3.0-9.5 fJ/bit/operation at room temperature using the cryocooling efficiency of 0.1%.
Abstract : The major objective of the project was to design and demonstrate operation of key components of a 30 GHz 16-bit RSFQ processor prototype implemented with the AIST/ISTEC 10 kA/cm sq. fabrication process. Our team has developed complete logical and physical designs of five RSFQ chips using the CONNECT cell library and RSFQ CAD tools developed at the Universities of Yokohama and Nagoya (Japan). The major results are the world's first successful design, fabrication, and demonstration of correct operation of a 20 GHz 8x8-bit parallel carry-save RSFQ multiplier with approximately 6K JJs, a 16-bit sparse-tree wave-pipelined RSFQ adder with approximately 10K JJs, and partial operation of an 8-bit ALU chip with approximately 9K JJs. The goal of the second phase of the project was to get detailed understanding of the performance, complexity, and energy efficiency of on-chip storage units implemented with superconductor Reciprocal Quantum Logic (RQL) using our RQL VHDL cell library tuned to the MIT Lincoln Laboratory 10 kA/cm2 248 nm process. The 8.5 GHz 1-4 Kbit 32-/64-bit multi-ported scratchpad memory, register files, write-through and write-back caches designed with RQL Non-Destructive Read-Out storage cells have the average energy consumption of 3.0-9.5 fJ/bit/operation at room temperature using the cryocooling efficiency of 0.1%.
New superconductor single flux quantum logics with no static power dissipation in bias resistors, such as Reciprocal Quantum Logic (RQL), offer opportunities to create energy-efficient superconductor processors operating at high frequencies with ultra-low power consumption. This paper discusses the results of our work on the cell-level design and analysis of a benchmark set of 32-/64-bit RQL processor integer and floating-point units such as adders, multipliers, an arithmetic-logic unit, and an array shifter, as well as small 1-4 Kbit RQL on-chip storage components such as register files, on-chip memory, and the top level caches. Our layout-aware design process includes the complete cell-level design and approximate physical layout of the circuits followed by the VHDL simulation, verification, and energy profiling using our RQL VHDL cell library tuned to the future MIT Lincoln Laboratory 10 kA/cm2 248 nm process with 10 Nb metal layers and the minimum JJ critical current of 38 μA. Our designs have the energy efficiency of ~1.0 single-precision TFLOPS/W and '0.5 double-precision TFLOPS/W for floating-point units, and ~1-24 TOPS/W for 32-bit integer units at room temperature using the cryocooling efficiency of 0.1 % (1000 W/W). The 1-4 Kbit 32-/64-bit multi-ported scratchpad memory, register files, write-through and write-back caches designed with RQL Non-Destructive Read-Out storage cells have the average energy consumption of 3.0-9.5 fJ/bit/operation at room temperature using the cryocooling efficiency of 0.1%. While these results are very promising, more work is needed to evaluate the contribution of the energy costs of instruction scheduling and off-chip main memory access to the energy efficiency of RQL computing across a whole system.
With Moore's law approaching its physical limits, low-temperature computing technology is ushering in unprecedented development opportunities. Rapid single-flux-quantum (RSFQ) circuit technology is currently the most mature superconducting integrated circuit technology. Based on the current fabrication process, we propose an 8-bit bit-parallel RSFQ microprocessor. The proposed microprocessor processes 8-bit data each clock cycle. Ten different instructions are executed. The microprocessor mainly consists of an on-chip instruction memory, two data registers, an instruction decoder, an 8-bit bit-parallel arithmetic logic unit, and a program counter. The microprocessor contains 7702 JJs (based on the Open Dataset of CONNECT Cell Library for AIST ADP2) without considering splitters, Josephson transmission lines, and passive transmission lines. We perform a logic-level simulation of the proposed microprocessor. The simulation results show correct operation with a target frequency of 16.7 GHz.
As part of the effort to develop a superconductive microprocessor, we constructed a rapid single flux quantum (RSFQ) 4-bit arithmetic and logic unit (ALU) in a pipelined structure. To speed up the circuit, we used a forward clocking scheme. The 4-bit ALU consisted of ten RSFQ Half Adders, four 2 /spl times/ 2 switches, and several D flip-flops. By commutating output ports of the half adder, we were able to produce AND, OR, XOR, and ADD functions. The 4-bit ALU was fabricated using the Korea Photonics Technology Institute's ten-level 1.0 kA/cm/sup 2/ Nb process. The size of the ALU was 3.0 mm /spl times/ 1.5 mm, fitting in a 5 mm /spl times/ 5 mm chip. The fabricated 1-bit ALU block was successfully tested at up to a 40 GHz clock frequency. The complete 4-bit ALU operated correctly at up to a 5 GHz clock frequency.
Superconductor single flux quantum (SFQ) technology is one of the promising candidates for energy-efficient high-performance computing. A new generation of SFQ logics, Reciprocal Quantum Logic (RQL) with no static power dissipation in bias resistors, offers an opportunity to dramatically decrease energy consumption in superconductor processors. Although several low complexity RQL processing units have already been demonstrated, the use of RQL for local storage design has not been explored yet. The objective of this dissertation is to design on-chip local storage structures such as memory, register files, and caches with RQL technology and analyze their energy efficiency, complexity, and performance characteristics. The physical chip design of these RQL storage units is not feasible at this point because both CAD tools for physical VLSI chip design and as well as a target fabrication process are under development at Northrop-Grumman Systems Corp. (Baltimore, MD) and MIT Lincoln Laboratory, respectively. In order to achieve our goal, the layout-aware cell-level design process using VHDL RQL cell library developed at the Ultra High Speed Computing Lab in Stony Brook University has been used. The SBU VHDL RQL cell library specifies the dynamic and stand-by energy consumption, latency, JJ complexity, and approximate sizes of individual cells based on the input received from the JJ-level RQL designers. Clock propagation skew and wire delays are accounted for during circuit simulation. The circuit simulation is done with Mentor Graphics design and verification tools. As a result of the work, key characteristics of the 8.5 GHz multiported RQL storage structures with their capacity in the range of 1-4 kbit have been determined. The average energy consumption of the RQL storage designs is ~3.0-9.5 fJ/bit/operation at room temperature and the cryo-cooling efficiency is 0.1%. The data from this dissertation also reveal the critical issues that need to be considered in the RQL storage design. These will be helpful for further development on superconductor VLSI design. | 124 pages
New superconductor single flux quantum (SFQ) technology, such as Reciprocal Quantum Logic (RQL), is currently considered one of the promising candidates for highperformance energy-efficient computing. This paper presents our work on the design and detailed energy efficiency analysis of three types of 32- and 64-bit RQL multi-ported pipelined local storage structures (13 total), namely 1) random access memory (RAM) and register files, 2) direct-mapped write-through and write-back caches, and 3) first-in-first-out (FIFO) buffers. Our layout-aware cell-level design process uses a VHDL RQL cell library developed at the Ultra High Speed Computing Laboratory at Stony Brook University (SBU). The SBU VHDL RQL cell library specifies the dynamic and standby energy consumption, gate delays, a number of Josephson junctions (JJs) per cell, and approximate sizes of individual cells based on the parameters of the 248 nm 100 μA/μm210 Nb metal layer SFQ fabrication process currently under development at the MIT Lincoln Laboratory. Gate and wire delays as well as clock skew are taken into account during digital circuit simulation done with Mentor Graphics CAD tools. After completing a physical chip layout, the circuit models need to be updated and re-simulated to include the effects of parasitic inductances and actual wire lengths on signal propagation delays. To meet both performance and energy efficiency targets, the RQL storage structures were designed with RQL non-destructive read-out single-bit storage cells. We chose a relatively moderate clock frequency of 8.5 GHz for all storage units to keep their read latencies in the range of 1- 3 cycles. The most complex design in terms of JJs is a tripleported 4 Kbit 64x64-bit register file with 253,918 JJs and its read access latency of 338 ps. The highest energy consumption in terms of energy/operation/bit (~9.5 aJ at 4.2 K) is for a write hit in a 2 Kbit 32-bit wide write-back cache. The average energy consumption of the RQL storage designs varies from ~1.6 aJ/operation/bit for a small 4x32-bit FIFO to 7.3 aJ/operation/bit for the 2 Kbit write-back cache at 4.2 K. Given the cryocooler efficiency of 0.1%, this means the energy consumption of ~1.6-7.3 fJ/operation/bit at room temperature. The physical implementation of the RQL storage units will become feasible upon the development of the target MIT fabrication process and CAD tools for VLSI RQL chip design in 2015-2016.
In this paper, we discuss the architecture, design, and testing of the first 16-bit asynchronous wave-pipelined sparse-tree superconductor rapid single flux quantum adder implemented using the ISTEC 10 kA/cm2ADP2.1 fabrication process. Compared to the Kogge-Stone adder, our parallel-prefix sparse-tree adder has better energy efficiency with significantly reduced complexity (at the expense of latency) and almost no decrease in operation frequency. The 16-bit adder core (without SFQ-to-dc and dc-to-SFQ converters) has 9941 Josephson junctions occupying an area of 8.5 mm2. It is designed for the target operation frequency of 30 GHz with the expected latency of 352 ps at the bias voltage of 2.5 mV. The adder chip was fabricated and successfully tested at low frequency for all test patterns with measured bias margins of +9.8%/-10.7%.