Beyond-CMOS roadmap—from Boolean logic to neuro-inspired computing

Abstract

As one of the chapters in the International Roadmap of Device and Systems (IRDS), the “Beyond CMOS (BC)” roadmap surveys and catalogs emerging devices and materials, and evaluate their potential and challenges gating their acceptance by the industry. While CMOS is expected to continue to dominate as the platform technology, beyond-CMOS devices may enable novel computing paradigms and efficient hardware accelerators to augment the CMOS platform. Emerging device-architecture co-design and co-optimization are important for achieving the efficiency and functionalities beyond the limit of CMOS technologies. This paper provides a brief overview of the IRDS BC chapter, with the emphasis on the opportunities of beyond-CMOS devices and architectures for neuro-inspired computing paradigms.

From fundamentals to frontiers: a review of memristor mechanisms, modeling and emerging applications

“…The changes of the 'slow' state variables are controlled by a driven adaptation function g(U,x) with the characteristic time t k . The main properties of the memristive systems are (a) highly nonlinear characteristics in function F, (b) Significant memory effect which implies that the current value of Y depends on history of the sample by its need to integrate eqn (2). 10 Because of the memory feature, there is a nonlinear relationship between current (I) and voltage (V) causing a pinched hysteresis loop.…”

“…1 However, as CMOS scaling approaches fundamental physical constraints, it is becoming increasingly difficult to continue improving computing performance using this technology. 2 To address the memory bottleneck of the von Neumann architecture, much research has been done to create resistive switching devices for in-memory computing (IMC) applications. These designs share the fundamental idea of co-locating memory and processing with biological systems, where neurons integrate input signals and then create output signals that are sent to downstream neurons via comparable synaptic connections.…”

Harnessing ferroic ordering in thin film devices for analog memory and neuromorphic computing applications down to deep cryogenic temperatures

“…The future of computing beyond Complementary Metal Oxide Semiconductor (CMOS) era requires extensive use of material and device physics to perform computation at the atomic level (Manipatruni, 2018;Chen, 2022). Development of neuromorphic computing hardware, that is devices with bio-plausible functionalities for implementing neural network operations in hardware, requires different kinds of volatile and non-volatile memories to implement synaptic and neuronal functionalities.…”

Recent Advances and Future Prospects for Memristive Materials, Devices, and Systems

“…Exponential scaling of the switching channel in CMOS devices, followed by Moore’s law, is one of the key enablers for the advancement . However, CMOS scaling is approaching fundamental physical limits, which has necessitated the need to look for alternative switching devices to continue improving computational performance. − An ionic-based resistive switching device (i.e., a memristor) is a leading candidate in this context. − Due to its programmable analog memory effect, which resembles the function of biological synapses in the human brain, memristors have been regarded as a building block of neuromorphic (i.e., brain-like) electronics. This technology has the potential not only to provide biomimetic devices that support human-like data processing but also for influencing even the most basic forms of electronics (Figure ).…”