Brain-inspired computing via memory device physics

Abstract

In our brain, information is exchanged among neurons in the form of spikes where both the space (which neuron fires) and time (when the neuron fires) contain relevant information. Every neuron is connected to other neurons by synapses, which are continuously created, updated, and stimulated to enable information processing and learning. Realizing the brain-like neuron/synapse network in silicon would enable artificial autonomous agents capable of learning, adaptation, and interaction with the environment. Toward this aim, the conventional microelectronic technology, which is based on complementary metal–oxide–semiconductor transistors and the von Neumann computing architecture, does not provide the desired energy efficiency and scaling potential. A generation of emerging memory devices, including resistive switching random access memory (RRAM) also known as the memristor, can offer a wealth of physics-enabled processing capabilities, including multiplication, integration, potentiation, depression, and time-decaying stimulation, which are suitable to recreate some of the fundamental phenomena of the human brain in silico. This work provides an overview about the status and the most recent updates on brain-inspired neuromorphic computing devices. After introducing the RRAM device technologies, we discuss the main computing functionalities of the human brain, including neuron integration and fire, dendritic filtering, and short- and long-term synaptic plasticity. For each of these processing functions, we discuss their proposed implementation in terms of materials, device structure, and brain-like characteristics. The rich device physics, the nano-scale integration, the tolerance to stochastic variations, and the ability to process information in situ make the emerging memory devices a promising technology for future brain-like hardware intelligence.

2022 roadmap on neuromorphic computing and engineering

“…An additional circuit challenge is the mixed analog-digital computation, which results in the need for large and energyhungry analog-digital converter circuits at the interface between the analog crosspoint array and the digital system. Finally, neuromorphic circuits seem to take the most benefit from hybrid integration, combining front-end CMOS technology with novel memory devices that can implement MVM and neuro-biological functions, such as spike integration, short-term memory, and synaptic plasticity [15]. Hybrid integration may also need to extend, in the long term, to alternative nanotechnology concepts, such as bottom-up nanowire networks [16], and alternative computing concepts, such as photonic [17] and even quantum computing [18], within a single system or even a single chip with 3D integration.…”

Nonideality‐Aware Training for Accurate and Robust Low‐Power Memristive Neural Networks

“…The models may incorporate principles like Fowler-Nordheim tunneling, thermionic emission, and Poole-Frenkel emission. [46] Data in Figure 2a,b were best fit using Poole-Frenkel model in Equation (9).…”

Nonideality-Aware Training for Accurate and Robust Low-Power Memristive Neural Networks

“…To construct an I-V curve from a given γ, one may assume that the equality in Equation ( 8) holds for all V , not just V ref . This would produce a relationship between current and voltage shown in Equation (9).…”

“…In this specific case, an alternative can be memristor-based ANNs, or memristive neural networks (MNNs). With this approach, memristive crossbar arrays are used to physically compute vector-matrix products which are one of the most fundamental operations in ANNs [8,9]. This is done without the need to constantly move large amounts of data: matrix entries are encoded into memristor conductances, vector entries-into applied voltages, and the result of the operation is extracted from the output currents produced according to Ohm's law and Kirchhoff's current law [8].…”