Probabilistic Computing with p-bits: Optimization, Machine Learning and Quantum Simulation

The increasing computational demands of the modern AI has been in steady conflict with the slowing down of Moore's Law driven era of electronics. To meet the demands of AI in the new era, augmenting existing complementary metal oxide semiconductor (CMOS) technology with emerging technologies has become a promising new direction. One research thread along this line is the notion of physical computing, where a set of computational tasks are mapped onto a programmable "computer" whose natural physics leads to desired outcomes for computational problems. What is different from traditional computers in this approach is a shift from the focus of general-purpose to domain-specific computing. One example is probabilistic computing with fluctuating probabilistic bits (p-bit) that are conceptually in between bits and qubits. Among many possible implementations, p-bits that use the inherent stochasticity of magnetic tunnel junctions is emerging as one of the most scalable, energy-efficient and manufacture-ready option and their scaled implementation could be of particular use in a variety of applications in combinatorial optimization, machine learning and quantum simulation. Below, I discuss the rapid pace of progress of this field from device, architecture and algorithmic perspectives.