In-memory photonic dot-product engine with electrically programmable weight banks

Researchers from the Bhaskaran Group, University of Exeter and Heidelberg University have worked together to achieve computation success to resolve the von-Neumann bottleneck for electronically reprogammable photonic circuits based on phase-change chalcogenides.  

This milestone is described in their paper 'In-memory photonic dot-product engine with electrically programmable weight banks' published in Nature Communications.  In the paper, the authors demonstrate an in-memory photonic-electronic dot-product engine which decouples electronic programming of phase-change materials (PCMs) and photonic computation.  They discuss how they developed non-volatile electronically reprogrammable PCM memory cells with a record-high 4-bit weight encoding, the lowest energy consumption per unit modulation depth (1.7 nJ/dB) for Erase operation (crystallisation) anf a high switching contrast (158.5%) using non-resonant silicon-on-insulator waveguide microheater devices.  

Utilising the above, they were able to perform parallel multiplications for image processing with a superior contrast-to-noise ratio (>87.36) which led to an enhanced computing accuracy (standard deviation o<0.007).  An in-memory hybrid computing system was developed in hardware for convolutional processing for recognising images from the MNIST database with inferencing accuracies of 86% and 87%.