LOw-power Digital deep leaRning Inference Chip

The LODRIC project is a follow-up project from "LO3ML", a project from the BMBF's Sprunginnovationswettbewerb „Energieeffizientes KI-System“, in which it won the "ASIC 130nm" track [1]. In this project, AI methods were to be used to detect the disease pattern for atrial fibrillation from ECG data, and to do so as energy-efficiently as possible.

The goal of the follow-up project LODRIC is now the development of a design methodology for energy-efficient digital AI chips with embedded non-volatile memory elements and their prototypical application on the basis of three different applications. In doing so, the main innovation of the "Lo3-ML" project from the "Energy Efficient AI System" leap innovation competition, namely the development of data-flow-oriented computer architectures in conjunction with distributed, non-volatile weight memory and strongly (ternary) quantized weights, is to be taken up and methodically developed further in a targeted manner.
Accordingly, two sub-goals can be formulated from the aforementioned overall goal:
1. development of a development framework which, using existing, configurable building blocks, creates an energy-efficient computer architecture using various optimization algorithms.
2. prototypical implementation of an optimized architecture as a real chip, generated according to the methodologies of the development framework, which can be used for three different applications (here: ECG data analysis, predictive maintenance and radar signal processing) can serve.

[1]: www.elektronikforschung.de/service/aktuelles/pilotinnovationswettbewerb