Hybrid electronic-photonic architectures for brain-inspired computing

Introduction

As artificial intelligence (AI) proliferates, hardware systems that can perform inference at ultralow latency, high precision, and low power are crucial and urgently required to deal especially quasi-locally, i.e., in the edge, with massive and heterogeneous data. These systems need to respond in real time and avoid unintended consequences while functioning in complex and often unpredictable environments.

Limitations of Conventional Systems

Conventional digital electronics and the associated computer architecture are unable to meet these stringent requirements with sub-ms latency inference and a sub-10W power budget. This is particularly evident when using convolutional neural networks (CNNs) on benchmarks such as ImageNet classification.

HYBRAIN's Vision

HYBRAIN's vision is to realize a pathway for a radical new technology with ultrafast (~1 microsecond) and energy-efficient (~1 watt) edge AI inference. This will be based on a world-first, brain-inspired hybrid architecture of integrated photonics and unconventional electronics.

Brain-Inspired Architecture

The deeply entwined memory and processing, akin to the mammalian brain, obviates the need to shuttle around synaptic weights. The most stringent latency bottleneck in CNNs is found in the initial convolution layers.

Innovative Approach

Our approach will take advantage of the ultrahigh throughput and low latency of photonic convolutional processors (PCPs) employing novel phase-change materials in these initial layers to radically speed up processing.

Processing Output

Their output will be processed using:

1. Cascaded electronic linear classifier layers
2. Nonlinear classifier layers, based on memristive (phase-change memory) crossbar arrays and dopant network processing units, respectively.

Collaborative Efforts

HYBRAIN's science-towards-technology breakthrough brings together the world's top research groups from academia and industry in complementary technology platforms. Each of these platforms is already highly promising, but by integrating them, HYBRAIN will have a transformative effect on overcoming existing barriers of latency and energy consumption.

Future Applications

This integration will enable a whole new spectrum of edge AI applications throughout society.