Heterogeneous Machine Intelligence

Heterogeneous intelligent systems bring together deep learning, neuromorphic models, brain-inspired associative memories, and fundamentally new ways of encoding information.

Heterogenous machine intelligence is all about “integration”: when the integration of diverse AI models meets the integration of semiconductor technologies, the resulting 3D heterogeneous intelligent systems lead a promising path towards the frontiers of future applications, where our next-generaiton electronic systems must be able to sense and process multi-modal information, while delivering insights on the fly, efficiently, autonomously, and securely.

There are plenty of opportunities by embracing the third dimension for our next-generation integrated systems: the up-scaling of system capacity and connectivity in 3D will open up a large space for application mapping. We embrace the vast design space offered by new physics, materials, and devices to build energy-efficient VLSI circuits. More importantly, the continuum of integration and packaging across 2D and 3D worlds will open up tremendous opportunities for the integration of nanotechnologies and chip fabrics as the foundation for 3D intelligent systems.

Our current directions include:

  • HW/SW modeling framework: DNN and non-DNN models, hardware technologies and system integration.
  • Architectural explorations, circuit design, tapeout, and integration.
  • Experimental demonstrations at various scales, from test structures, to chips, and to chiplets.