The case for hybrid analog neuromorphic chips based on silicon and 2D materials

We discuss two closely related technologies for the fabrication of analog neuromorphic circuits, one based on a standard silicon CMOS process and a second one based on heterostructures of two-dimensional materials (2DMs). We analyze the relative advantages and disadvantages of such solutions and the case for a hybrid technology, that can leverage the advantages of both. We show that floating-gate FETs - both in silicon and 2DMs - are very promising as analog non-volatile memories (NVMs) enabling good analog computing precision, which is a distinct advantage with respect to other proposed NVMs. We show that the challenge of equipping devices at the edge of the cloud with cognitive capabilities can be addressed using large analog blocks in machine learning chips and innovations in architectures, circuits, and technology.