SKM 2021 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

SYNC: Symposium Advanced neuromorphic computing hardware: Towards efficient machine learning

SYNC 1: Symposium: Advanced neuromorphic computing hardware: Towards efficient machine learning

SYNC 1.5: Invited Talk

Wednesday, September 29, 2021, 12:15–12:45, Audimax 1

In-memory computing with non-volatile analog devices for machine learning applications — •John Paul Strachan — Peter Grünberg Institute (PGI-14), Forschungszentrum Jülich GmbH, Jülich, Germany — RWTH Aachen University, Aachen, Germany

I describe our work to build non-von Neumann computing systems for machine learning and other computing applications. We are able to improve speed and power by leveraging emerging non-volatile and analog devices (e.g., memristors) and combining with mature CMOS technology, enabling the construction of novel circuits and architectures. We describe the acceleration of linear algebra operations and also complex pattern storage and retrieval, which are core operations in modern deep learning and broader machine learning workloads. We also build improved Content Addressable Memory (CAM) circuits that can be used in a variety of computing applications from network security, genomics, and many types of data classification. We forecast significant improvement over CPUs, GPUs, and custom ASICs using these new architectures. I will also describe work in addressing the types of errors often observed in analog systems, both in mitigating their effects as well as harnessing them productively.