Cell library is the keystone component that enables adoption of advanced electronic design automation (EDA) tools, such as logic synthesis and automatic place-and-route. The EDA tools are essential ...for scaling circuit complexity by orders of magnitude. We have designed a dual RSFQ/ERSFQ cell library for the MIT-LL SFQ5ee process, that can be used with the superconductor EDA tools suite that is being developed. In addition to satisfying the margins criterion, the performance of each cell has been optimized for Monte-Carlo statistical variations across multiple process corners including minimizing the spread of timing distributions. To enable a digital design flow using HDL simulations with timing back-annotation Liberty files have been developed for multiple process corners, using the load-dependent timing char-acterization. The cells have been designed for a standard height of 40 μm with a grid size of 20 μm. The library provides dedicated tracks for signal and power routing. Multiple independent biases are supported for RSFQ designs. The cells can be interconnected either by abutting or using passive transmission lines. Dedicated moat slots have been provided which are uniformly distributed across the cell. All cells are re-optimized post-layout. The library currently contains 22 unique types of cells. Initial validation of the cell library was performed by designing RSFQ and ERSFQ shift registers for the MIT-LL SFQ5ee fabrication process, which yielded wide operating margins. In addition, we present measurement results for a chip designed and fabricated to characterize several library cells using a multiplexing scheme.
In this paper, we designed and demonstrated interface circuits between AQFP (adiabatic quantum-flux-parametron) and RSFQ (rapid single-flux-quantum) circuits according to the established AQFP and ...RSFQ cell libraries using the MIT LL SFQ5ee process. The principles of these interfaces are based on ones that were previously designed and demonstrated using a four-metal-layer superconductor process. The AQFP-to-RSFQ interface is based on an RSFQ comparator, where an output current from an AQFP buffer is converted to an SFQ pulse output. The RSFQ-to-AQFP interface is based on an RSFQ D flip-flop, where the internal state of the D flip-flop is read out by an AQFP buffer. Thanks to the eight-metal-layer process, the interfaces can be designed with reduced circuit area compared to original designs with the four-metal-layer process. We designed, fabricated, and tested an AQFP-RSFQ hybrid multiplexer test circuit as an example of utilization of the interfaces, which results in correct operation with reasonably wide bias margins in low-frequency tests.
In design and layout of rapid single flux quantum circuits, multiple independent biases are often desired to provide flexibility in timings of critical paths. In absence of automated power routing, ...power nets are often included as part of the layout cells. Multiple variants of each cell need to be manually created to support different biasing configurations. The parametric approach for routing power nets provides the flexibility to dynamically change biasing configurations, based on the same parametric cell. Multiple cells in a circuit block can be configured to be connected to a single power net or can be biased using several independent power nets. With the goal of minimizing the cell size, it is often difficult to maintain rotational symmetry for the power and passive transmission line (PTL) tracks. The parametric approach also enables rotating the cell while avoiding misalignment of the tracks. In addition, our implementation can dynamically add moat bridges to connect ground planes isolated by moats. This is especially useful to bridge very long moats that span two or more adjacent cells. Furthermore, this approach of dedicated tracks for power and PTL routing is also more amenable to design automation.
Performing comprehensive timing characterization of library cells and establishing their model-to-hardware correlation are essential for building robust integrated circuits and scaling circuit ...complexity. To measure the propagation delay of fabricated library cells, we have designed a differential delay measurement scheme using SFQ ring oscillators. The reference and design-under-test (DUT) rings are designed using circuit elements that are either common to both rings or are of nominally similar design, except for the DUT. By inserting a single pulse in the ring, the oscillation period is measured using two methods: (1) direct voltage measurement, and (2) frequency measurement using a frequency divider followed by an SFQ-to-DC converter. Subtracting delay of the two rings gives us the propagation delay of the DUT. A 1 cm by 1 cm chip and another 5 mm by 5 mm chip, incorporating stand-alone ring oscillator test structures for different library cells, have been fabricated in the MIT-LL SFQ5ee process. The chips also contain test circuits to validate two variants of the input/output multiplexing scheme that facilitate reducing the pads requirement. Two copies of a chip with fifteen DUTs were tested. The measured and simulated results for the voltage and frequency of the ring are compared. The simulations account for critical current density ( J c ) and sheet resistance (Rs) fabrication parameters. In addition, we have used the J c as a single free parameter to get a better matching with measurement results. For one of the chips, a 15% higher J c in simulations results in better matching with measured frequency results for all the DUTs. For example, the discrepancy between simulated and measured DUT ring frequency for the NOT cell was reduced from 19% to 1% when the J c was changed from nominal to 15% higher in simulations. Detailed measurement and simulation results for several DUTs are presented.
Arithmetic Logic Unit (ALU) is an integral part of digital signal processing applications and computing systems. We used ALU, based on Kogge-Stone adder, as a reference circuit to experimentally ...validate the advanced design flow and the dual RSFQ/ERSFQ standard cell library. Using the advanced design flow, the ALU sub-blocks were optimized across multiple process corners using Monte-Carlo simulations incorporating statistical variations in the process parameters. The correct operation of ALU was verified in the digital HDL simulation using static timing analysis pre-fabrication. We designed the RSFQ 1-bit and 4-bit ALU using the standard cell library approach for MIT-LL 100 µA/µm 2 SFQ5ee fab node. For the ERSFQ 1-bit ALU, the bias current overhead for the feeding JTLs was designed to be 100% of the ERSFQ DUT bias current. The RSFQ 1-bit ALU, designed for low frequency functional testing, demonstrated greater than ±15% bias margins. We demonstrated successful operation of the RSFQ 4-bit ALU with greater than ±6% bias margins at 20 GHz, greater than ±5% bias margins at 40 GHz, and greater than ±4% bias margins at 50 GHz clock frequency with BER less than 10 -12 . The ERSFQ 1-bit ALU worked up to 30.72 GHz clock frequency with BER less than 10 -12 . In addition, we analyzed the model-to-hardware correlation for the INIT sub-block of the ALU. The simulations accounted for critical current density ( J c ) and sheet resistance ( R s ) process parameters. We noted the discrepancy in simulated and measured margins and studied that a 15% higher J c in simulations results in better model-to-hardware correlation for the INIT sub-block.
Abstract The semiconductor industry seeks energy-efficient alternatives as Moore’s law nears its limits. The Single Flux Quantum (SFQ) integrated circuits (ICs) using thousands of niobium Josephson ...junctions (JJs) and operating at 4 K show great promise for digital computing circuits at high speed (>20 GHz) and low power (a few nW per junction). The leading logic families are Rapid Single Flux Quantum (RSFQ), and its energy-efficient variant (ERSFQ). IARPA’s SuperTools program aims to develop integrated design tools for superconductor electronics, targeting SFQ and Adiabatic Quantum-Flux-Parametron (AQFP) logic families. This paper presents a passive transmission line (PTL) based standard cell library for SFQ logic, designed with Synopsys Electronic Design Automation (EDA) software tools for MIT-LL 100μA/μm 2 SFQ5ee fab node. The dual RSFQ/ERSFQ standard cell library facilitates seamless integration of SFQ RTL-to-GDS design flow with Synopsys Fusion Compiler, an automated design tool. The SFQ RTL-to-GDS flow entails logic synthesis, checking, placement, clock synthesis, and routing. Row-based placement for library cells and H-tree clock tree structures are employed. Fusion Compiler’s effectiveness is validated with Hypres designs such as finite impulse response (FIR) filters, scalable multiply-accumulate (MAC) units, and memory arrays, comparing single and dual clocking schemes. The synergy between Hypres and Synopsys achieves a milestone by demonstrating the design of a digital superconducting circuit with over 10 million JJs, facilitated by a fully automated design tool for the first time. Challenges in very large-scale SFQ scaling are also discussed.
We have developed a microscopic tunneling Josephson junction compact model (TJM) framework and incorporated this in the Synopsys HSPICE commercial circuit simulator. Our model follows the Odintsov, ...Semenov, and Zorin (OSZ) approach, whereby tunnel current amplitude tables are reduced to a fitting parameter set, which is accessed by the simulator compact model. In addition, our model takes into account the computed BCS energy gap as a function of temperature and uses a table-lookup technique for rapid generation of temperature-specific parameters, enabling efficient temperature sweep and Monte-Carlo variation. This avoids the rather long computations required for generation of fit parameters for specific temperatures. Initially we will review the algorithms and features of our TJM model, then we will report simulation results of our model focusing on comparison to results using the standard RSJ model.
Passive transmission lines (PTLs) provide an energy-efficient means of transporting pulsed single flux quantum (SFQ) signals between logic gates and blocks on a chip, or even between chips through a ...multi-chip module carrier. Although functional up to 70 GHz, our previous designs manifested sharp, length-dependent resonances resulting in diminished operating margins. These are particularly inconvenient since the resonant length is about 1 mm at 40-50 GHz, where most SFQ circuits currently function. To avoid the need to keep the PTLs < 1 mm in length, we have optimized the PTL driver-receiver pairs to sufficiently suppress these resonances; this helps facilitate the use of automated routing tools. We present experimental results, in the 10-70 GHz frequency range, for different driver-receiver pairs designed for PTLs in our dual RSFQ-ERSFQ cell library. This cell library, targeting MIT Lincoln Laboratory's SFQ5ee fabrication process, uses PTLs with two types of ground plane configuration. The first variant has signal in M1 and/or M3 with symmetrical ground planes in M0, M2 and/or M2, M4 respectively. The second variant has signal in M2 and/or M3 with asymmetrical ground planes in M1 and M4. We also present a comparison with wider, 4 Ω, PTLs that are more suitable for transport between distant logic blocks. Next, we present experimental results of PTL lengths ranging from 1.2-12 mm over the same 10-70 GHz frequency range. Finally, we report on the design of circuits for investigating crosstalk between two PTLs and their measurement results.
Single-flux quantum (SFQ) logic based high-speedperiodic-threshold flash converter circuits require multiple non-linear and correlated parameters tuned precisely to function optimally. These ...parameters cannot be pre-determined from simulations and change with the clock frequency, thus requiring manual optimization every time the clock frequency is changed. In this work, we demonstrate automated bias optimization of an 8-bit superconducting SFQ flash ADC for the first time. A closed-loop 4X lower computational complexity hybrid particle-swarm-gradient-descent-optimization is demonstrated that first uses particle swarm optimization (PSO) to coarse tune the ADC biases and then applies gradient descent (GD) optimization for fine tuning. This results in a 12X reduction in the blind calibration time from several days manually, to a few hours using the proposed optimization scheme, with a resultant performance 2 dB better than an optimization done by a highly skilled human.
As the complexity of superconductor circuits grows, we envision a dense network of passive transmission lines (PTLs) being used to interconnect cells in rapid-single-flux-quantum (RSFQ) circuits. In ...our library approach, each cell has dedicated tracks as place holders for routing PTLs. Higher impedance PTLs are desirable due to their narrower width. However, at higher impedance, the margins for the PTL receiver degrade rapidly. We have designed, fabricated, tested, and simulated passive transmission lines (PTLs) for high-speed interconnects in the MIT-LL SFQ5ee 10 kA/cm 2 process. For the symmetric dual ground planes case, PTLs are in the M1 layer with M0 and M2 ground planes, or in the M3 layer with M2 and M4 ground planes. For the asymmetrical dual ground planes case, PTLs are in M2 or M3 layers with M1 and M4 ground planes. We report ±30% margins for these PTLs. We investigate the receiver margins for these PTLs and report impact of margins as a function of corners, interlayer transitions, variants of drivers, and receivers. We demonstrate 100 GHz PTL operation in ring oscillator measurements. We have also adopted multi-layer multi-conductor transmission line models for PTL simulation. We observe good model-to-hardware correlation for low- and high-frequency operation.