• Achieving a 3x improvement while optimized the front-end part of spatial cross-attention in BevFormer, compared to the latest tool chain.

• Enhanced 50x performance of BevPool from customer’s onnx network to our chip.

• Spearheaded modeling of Statistics & Matrix Multiply datapath and collaborated closely with RTL designer.

• Developed the Motion Compensated Temporal filtering compression/decompression modeling and validation

• Maintained video output module on computer vision chips.

• Validated the data path of the image digital signal processor on the latest computer vision chip.

• Developed and maintained the performance optimization and evaluation tool for convolution operation.

• Introduced a fast 3D convolution algorithm using tensor decomposition for the in-house vector proces-sor.

• Using Python to develop an automated debugging tool to assist with chip testing.

• Initiated an automated performance profiling tool in Python for on-chip task profiling.

• Customized and integrated ARM CoreSight debug systems on 6 products.

• Consulted for ARM CoreSight debug systems as an internal expert.

• Successfully delivered ARM Cortex-A53 Octa-core integration by internal user’s request.

• Maintained 4 peripheral IPs on 10+ products.