We develop physical and statistical models to understand how probabilistic computers behave and to guide the design of new devices and circuits. Our work connects spin transport and magnetization dynamics to the collective behavior of interacting stochastic variables. We ask which features of the physics a model must retain and how its approximations affect the computation.
Models for devices and circuits
Our foundation is modular spin-circuit modeling. We represent spin transport and magnetization dynamics as connected circuit modules that can be combined with transistor models in standard circuit simulators. This lets us examine how a device behaves inside a larger circuit, including electrical loading and its effects on energy and delay. We compare models with established theory and experiments before using them to explore new structures.
Beyond binary states
Our Heisenberg-machine studies connect the dynamics of continuous spin vectors to statistical models with programmable interactions. In collaboration with experimental groups, we also use analytical models and spin-circuit simulations to interpret electrically controlled vector spin injection in graphene.
Correlations and constraints
At the network level, we study how approximations change stochastic dynamics. Cluster mean-field theory retains interactions within each cluster while replacing interactions across its boundary with effective fields. Changing the partition allows different correlations to be represented directly. For multi-state Potts models, local choices can be represented directly, while mean-field feedback provides an approximate way to impose global constraints without dense pairwise connections.
Recent Publications
- Extreme-Scale Ising Machines with Cluster Mean-Field Theory. Preprint (2026).
- Spin Vector Control for Heisenberg-Inspired Probabilistic Computing. Preprint (2026).
- Restoring Sparsity in Potts Machines via Mean-Field Constraints. npj Unconventional Computing (2026).
- Connecting physics to systems with modular spin-circuits. npj Spintronics (2024).
- Heisenberg machines with programmable spin circuits. Physical Review Applied (2024).
Foundational Publications
-
T. Y. T. Hung, K. Y. Camsari, S. Zhang, P. Upadhyaya, Z. Chen, Direct Observation of Valley Coupled Topological Current in MoS2, Science Advances, Vol. 5, 4 (2019)
-
M. M. Torunbalci, P. Upadhyaya, S. A. Bhave, K. Y. Camsari, Modular Compact Modeling of MTJ Devices, IEEE Transactions on Electron Devices, Vol. 65, 10, 4628-4634 (2018)
-
K. Y. Camsari, S. Ganguly, D. Datta, S. Datta, Non-Equilibrium Green’s Function based Circuit Models for Coherent Spin Devices, IEEE Transactions on Nanotechnology Vol. 18, 858-865 (2019)
-
K. Y. Camsari, S. Ganguly, Deepanjan Datta, A modular spin-circuit model for magnetic tunnel junction devices, 2016 74th Annual Device Research Conference (DRC) (2016)
-
K. Y. Camsari, S. Ganguly, S. Datta, Modular approach to spintronics, Scientific Reports 5, 10571 (2015)
- K. Y. Camsari, S. Ganguly, D. Datta, S. Datta, Physics-based factorization of magnetic tunnel junctions for modeling and circuit simulation, 2014 IEEE International Electron Devices Meeting (IEDM) (2014)