Department of Electrical and Computer Engineering Ph.D. Public Defense

Design Methodologies for Single Flux Quantum VLSI Circuits

Gleb Krylov

Supervised by Professor Eby Friedman

Friday, July 2, 2021
1 p.m.

https://rochester.zoom.us/j/97400408689?pwd=cHBiT2s3ZXNwbmd6TW5LYW5rQXpKdz09
Meeting ID: 974 0040 8689
Passcode: 467997

Abstract:

Rapid single flux quantum (RSFQ), the most widely adopted type of superconductive digital logic, is a beyond-CMOS technology particularly appropriate for large scale, efficient, stationary computing such as supercomputers and data centers. The fabrication capabilities of modern superconductive foundries currently approach a million logic gates per IC; the complexity of practical systems, however, does not currently exceed several thousand gates. This “design gap” exists due to the lack of efficient computer aided design tools; in particular, methodologies aware of the issues inherent to very large scale integrated (VLSI) superconductive circuits.

In this dissertation, issues and solutions to enable VLSI RSFQ circuits and systems are presented. Topics such as memory, synchronization, bias networks, and testability are described, and models, circuits, algorithms, and design methodologies are proposed.

A simplified compact model of a superconductor-ferromagnetic transistor (SFT) is developed, to enable simulation and analysis of SFT-based memory circuits. A sense amplifier topology is also presented to read memory cells composed of magnetic tunnel junctions and nanocryotrons, providing a flexible and area efficient solution for reading spin-based memory. A globally asynchronous, locally synchronous clocking scheme and a network-on-chip topology are described, where the ambiguity of clock and data in RSFQ technology is exploited. Moreover, asynchronous dynamic SFQ (DSFQ) majority gates are also proposed. The use of asynchronous logic gates simplifies the clock network, while majority gates reduce the logic depth, enhancing performance. Synthesis optimization techniques to increase the performance of DSFQ circuits are presented. A methodology for the distributed placement of bias structures in large scale energy efficient RSFQ circuits is proposed, enabling precise control of the parasitic inductances within a bias network while also reducing area and power. A methodology for partitioning RSFQ circuits during the placement process is also described to enable current recycling and reduce the total current, complexity, and inductive noise coupling in the bias network. Sources of parasitic inductive coupling in superconductive circuits are characterized and mitigation solutions are proposed. Test point insertion and set/scan chain techniques compatible with RSFQ logic are proposed, and circuit techniques to reduce the area required by these test structures are described.