Synaptic Resistors for Concurrent Inference and Learning with High Energy Efficiency

Abstract

Transistor-based circuits with parallel computing architectures and distributed memories, such as graphics processing units (GPUs) from Nvidia, [9] tensor processing units (TPUs) from Google, [3,10] field-programmable gate arrays (FPGAs) from Intel, [11] and the TrueNorth neuromorphic circuit from IBM [12] have been developed to improve their energy efficiencies ( Figure 1a) to the range of 10 10 − 10 11 FLOPS W −1 (floating point operations per second per watt) by increasing parallelism and reducing global data transmission. However, their energy efficiencies are fundamentally limited by the energy consumptions on memory (≈10 −15 J bit −1 ) and signal transitions (≈10 −11 J bit −1 ) in digital computing circuits. [5,6] When transistors approach the limitations of their minimal sizes near the end of Moore's law, the energy efficiencies of transistorbased computing circuits are asymptotically saturated [4][5][6]13,14] (Figure 1a). Meanwhile, the information industry generates "big data" with exponentially increasing volumes, and leads to exponentially increasing power requirements for computations. [4,6,14] This trajectory is unsustainable as it would exceed the entire global power production in one or two decades [15] (Figure 1a). It is imperative to develop a new platform to facilitate inference and learning from "big data" in emerging intelligent systems with significantly higher energy efficiency than that of the transistor-based Turing computing platform.The human brain performs inference and learning from "big data" with an estimated speed (≈10 16 FLOPS) [16] comparable to the speed (≈10 17 FLOPS) of the fastest supercomputer, Summit, [17] but consumes much less power (≈20 W) than the supercomputer (≈10 7 W), and is much more energy-efficient (≈10 15 FLOPS W −1 ) than the supercomputer (≈10 10 FLOPS W −1 , Figure 1a). By contrast, the human brain concurrently performs spatiotemporal inference and learning in analog parallel mode [16,18,19] (Figure 1c) via a network of neurons connected by ≈10 14 synapses (Figure 1d). For inference, a wave of voltage pulses, V t m ( ) i Adv.

Light‐Stimulated Artificial Synapses Based on 2D Organic Field‐Effect Transistors

“…The human brain, consisting of 10 11 neurons and 10 15 synapses, can perform learning, calculating, thinking, and inference from the "big data" with an estimated speed of 10 16 floating point operations per second, which is comparable to the operating speed of the fastest supercomputer, Summit. [1,2] But, the energy consumption of the brain (20 W) is only about one millionth of that of Summit (10 7 W). The high efficiency of the brain has inspired scientists all over the world to mimic artificial synapses mainly focus on inorganic materials.…”

“…Herein, we report a new type of 2D OSC based synaptic transistor prepared by simple solution epitaxy [52,53] at room temperature. The p-type small molecule 2,7-dioctyl [1] benzothieno[3,2-b] [1] benzothiophene (C8-BTBT) was selected as the semiconductor material because of its high crystallinity and good solubility. The advantage of solution epitaxy is that 2D OSC films grow directly on the smooth water surface, which eliminates the effect of dielectric surface on the morphology of the deposited OSC film.…”

Chirality‐Enriched Carbon Nanotubes for Next‐Generation Computing

“…This compatibility was further confirmed by fabricating a 4 × 2 crossbar circuit using 72 synstor and 2 integrate-and-fire neuron circuits. Notably, this system shows a computational efficiency of ≈1.6 × 10 17 FLOPS W −1 that surpasses state-of-the-art supercomputers, [140] thus highlighting the potential of SWCNT electronics for next-generation neuromorphic hardware compared to other emerging materials (e.g., MoS 2 , perovskites, chalcogenides). [137]…”

Interface Modification in Three‐Terminal Organic Memory and Synaptic Device

“…With continuous attention, neuromorphic computing witnesses rapid development in many aspects, such as image processing [132,133] and speech signals recognition. [26] The human brain consists of around 10 11 neurons and 10 15 synapses, [131] which are connected with each other to form the complicated neural networks to process message coming into the brain with super-high efficiency. The contact part of two neurons is synapse, the basic unit to form the neural network.…”