Embedded Files
Ever-expanding connectivity nourished by high-performance digital computing brings both opportunities and threats. The achievements of deep neural networks (DNNs) in various fields promises immense potentials in IoT applications. However, the requirement of a huge amount of computations and parameters prevents the DNN models from directly deploying in always-on systems where the amount of resources is severely limited. This problem can be addressed by developing a hardware accelerator with an optimal dataflow for compact models or by adopting an energy-efficient hardwiring for application-specific targets. The growth of IoT also presents plenty of challenges in security. Physically unclonable function (PUF) and true random number generator (TRNG) have emerged to be promising for embedding hardware security in IoT devices. Our research includes
- Neuromorphic in-memory computing
- Deep learning accelerator
- Physically unclonable function
<Scalable-precision arbitrary quantization accelerator>
<Scalable-precision arbitrary quantization accelerator>
<On-chip Learning Neuromorphic Processor>
Arbitrarily Quantized ML Accelerator
[ISSCC 23]
Sparsity-Aware Dynamic-Precision DL Processor
[VLSI 23]
Cryptographic Wake-up Receiver
[VLSI 23]
Post-Quantum Cryptography Processor
[ESSERC 24]
<Selected Chip Implementations>
Page updated
Google Sites
Report abuse