Principal Investigators: Luqiao Liu, Marc Baldo

 

Networks formed by devices with intrinsic stochastic switching properties can be used to build Boltzmann machine, which has great efficiencies compared with traditional von Neumann architecture for cognitive computing due to the benefit from statistical mechanics of building blocks. In Boltzmann machine, each node within the network can switch to one of the two energy states with the switching probability determined stochastically through its interactions with other nodes within the network. This collective behavior of the interactive networks serves as an ideal platform for addressing many optimization and machine learning problems. Spin torque magnetic tunnel junctions are ideal candidates for realizing these Boltzmann networks due to its great tunability, small footprint, fast speed as well as their statistical-physics-governed stochastic switching. In this project, researchers will build thermally metastable three terminal magnetic tunnel junction devices for realizing these networks. The switching probability of each device – magnetic tunnel junction is determined by the states of other devices within the network through programmable weight factors that describe the interaction strength among different nodes. By integrating stochastic tunnel junctions and programmable resistive devices, the researchers will build a scalable Boltzmann machine with ultralow power consumption and fast speed.