Fine-Grained Transistor-Level QDI Asynchronous Crossbar Switch

Abstract

Neuromorphic systems have gained significant attention in recent years due to their potential to achieve energy-efficient and high-performance computing for a wide range of applications, including image and speech recognition, autonomous driving, and robotics. A critical requirement for these systems is high-radix crossbars, which enable efficient communication between neurons. However, designing high-radix crossbars poses significant challenges. To address these challenges, this paper proposes a fine-grained approach that utilizes (Quasi-Delay-Insensitive) QDI asynchronous circuits to design efficient high-radix non-blocking crossbars. The proposed approach is demonstrated through the design and post-layout simulation of a 32-bit 16x16 crossbar in 65nm CMOS at 1.2V. The results show that the proposed design achieves latency of 1.3ns and energy per bit of 9.26pJ, with substantial decrease in area of 0.1mm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. These results demonstrate the potential of the proposed approach to enable efficient and high-performance neuromorphic computing systems for a wide range of applications.