Scalable Pitch-Constrained Neural Processing Unit for 3D Integration with Event-Based Imagers

Maxence Bouvier (1) , Alexandre Valentian (1) , Gilles Sicard (2)

1 LSTA - Laboratoire Systèmes-sur-puce et Technologies Avancées
2 DOPT - Département d'Optronique

Event-based imagers are bio-inspired sensors presenting intrinsic High Dynamic Range and High Acquisition Speed properties. However, noisy pixels and asynchronous readout result in poor energyefficiency and excessively large output data rates. In this work, we use Convolutional Spiking Neural Network filters to compensate these drawbacks and reduce output bandwidth by 10x. We designed a neuromorphic core as a distributable block that benefits from 3D integration technology with direct and parallel access to 32x32 pixels, enabling reduced frequency operation. Post-layout simulations depict a peak energy efficiency with 2.83pJ per Synaptic Operation (equivalent to 0.093fJ/event/pix) at the nominal literature input event rate.

Mots clés: Spiking Neural Networks; Neuromorphic; 3D IC Technology; Near-Sensor Computing; Event-Based Vision

Domaines: Systèmes embarqués; Micro et nanotechnologies/Microélectronique

Maxence Bouvier, Alexandre Valentian, Gilles Sicard. Scalable Pitch-Constrained Neural Processing Unit for 3D Integration with Event-Based Imagers. 2021 58th ACM/IEEE Design Automation Conference (DAC), Dec 2021, San Francisco, France. ⟨hal-03470557⟩