Software

CrossSim is a GPU-accelerated accuracy simulator and co-design tool for analog in-memory computing. It models how hardware non-idealities in resistive crossbar arrays — programming errors, conductance drift, read noise, ADC quantization, and parasitic resistance — affect algorithm accuracy across neural network inference, signal processing, and linear algebra workloads. CrossSim provides a NumPy-like API, interfaces for PyTorch and Keras models, and supports hardware-aware training.

The Structural Simulation Toolkit (SST) is a modular parallel simulation framework for exploring innovations in highly concurrent systems, including processors, memory hierarchies, and network interconnects. Within SST Elements, I maintain Golem, a simulation model for analog accelerator tiles, and contribute to Carcosa, a component for modeling heterogeneous compute architectures.