Due to the slowing of technology scaling and architectures with separate microprocessors and memories, energy efficiency has become one of the major challenges for the conventional digital computing in the big data era. New computing schemes, such as neuromorphic computing and approximate computing, offer great opportunities for applications where energy consumption is a main concern. Such applications range from mobile platforms to server farms.
Various novel materials and devices have been proposed as potential candidates to replace or complement silicon in the post-Moore era. These emerging technologies -- both logic and memory -- have shown superior performance and unique properties than the traditional silicon technology.
However, material quality and fabrication processes for these devices are currently suffer from low device/ circuit yields due to imperfections. This means there is still much to be done before these emerging technologies are ready for scale use in conventional systems.
The only alternative is to develop error-resilient computing applications. To that end, starting with low-dimensional-materials (e.g. carbon nanotube, graphene, black phosphorous) based logic devices and emerging non-volatile memories (e.g. RRAM, MRAM), we will explore the impact of technology imperfection across the boundaries of device, circuit and system levels. We will also develop systematic methodology to improve circuit-level yield through device-circuit interactive design and optimization. We are actively exploring areas such as optimization framework which links material imperfection and variation with system-level yield for technology evaluation and projection, systematic design methodology for technology with process imperfection ready to be integrated into mainstream electronic design automation (EDA), and error-resilient circuit prototype with yield enhancement design.