CMOS + stochastic nanomagnets: heterogeneous computers for probabilistic inference and learning

With the slowing down of Moore's law, augmenting complementary-metal-oxide semiconductor (CMOS) transistors with emerging nanotechnologies (X) is becoming increasingly important. In this paper, we demonstrate how stochastic magnetic tunnel junction (sMTJ)-based probabilistic bits, or p-bits, can be combined with versatile Field Programmable Gate Arrays (FPGA) to design an energy-efficient, heterogeneous CMOS + X (X = sMTJ) prototype. Our heterogeneous computer successfully performs probabilistic inference and asynchronous Boltzmann learning despite device-to-device variations in sMTJs. A comprehensive comparison using a CMOS predictive process design kit (PDK) reveals that digital CMOS-based p-bits emulating high-quality randomness use over 10,000 transistors with the energy per generated random number being roughly two orders of magnitude greater than the sMTJ-based p-bits that dissipate only 2 fJ. Scaled and integrated versions of our approach can significantly advance probabilistic computing and its applications in various domains, including probabilistic machine learning, optimization, and quantum simulation.