Research survey

Prof. Shi joined Tsinghua University in 2013 and founded the Center for Brain Inspired Computing Research (CBICR) in 2014. His research covers the full stack of brain-inspired computing, spanning fundamental theories, neuromorphic devices, chips, software, systems and applications. Aiming to break von Neumann architecture limits and advance artificial general intelligence (AGI), his team develops core theories and technologies for brain-inspired models, algorithms, chips and computers.

In Nov 2015, the first heterogeneous cross-modal neuromorphic chip Tianjic was taped out for large-scale neural network simulation with high speed, real-time processing and low cost; relevant work was published in Science Robotics in Dec 2016.

In Oct 2017, the second-generation Tianjic chip adopted 28 nm process with 10 million synapses and 40,000 neurons, supporting both ANN and SNN including CNN, MLP and LSTM. It outperforms IBM TrueNorth in density, speed and bandwidth. The team also built a dedicated software toolchain for automatic algorithm compilation from mainstream ML frameworks and China’s first brain-inspired computing demo platform.

In 2018, two key algorithms were proposed: WAGE low-bit quantization framework enabling on-chip online learning, and STBP spatio-temporal backpropagation to resolve SNN non-differentiability and support deep SNN training via standard backpropagation.

In 2019, Gen-2 Tianjic completed system verification on autonomous unmanned bicycles, realizing parallel ANN-SNN computation and eliminating cross-network compatibility barriers. The result featured as a Nature cover story and was selected among China’s Top 10 Scientific and Technological Progress of 2019.

In 2020, the team pioneered the brain-inspired computing completeness theory, establishing a decoupled layered architecture for algorithms, hardware and systems to resolve bottlenecks of general neuromorphic platforms. The software toolchain was upgraded for end-to-end hybrid ANN-SNN deployment.

In 2021, a full-stack edge-oriented brain-inspired software platform was launched to unify heterogeneous chip development workflows, enabling one-click model migration and lowering industrial application thresholds.

Starting from 2022, the group researched complementary bionic visual perception and designed prototypes for primitive-representation vision chips with dual-channel retinal simulation. In 2023, Tianmou Core, the world’s first complementary brain-inspired vision chip, finished tape-out. Its dual-path bionic architecture cuts visual feature extraction power drastically with ultra-low-latency dynamic perception, accompanied by a full visual algorithm library.

In 2024, Tianmou Core won the Basic Research Leading Science and Technology Award at the World Internet Conference and appeared as a Nature cover article. The achievement was listed in China’s Major S&T & Engineering Progress and Top 10 Annual S&T News, while the team claimed the First Prize of Beijing Natural Science Award for heterogeneous fusion brain-inspired computing theories.

In 2025, the team systematically summarized the complete theoretical framework and core technologies of brain-inspired computing and perception chips, winning the Outstanding Achievement Award at the 2nd Zu Chongzhi Award for AI Frontier Innovation. The Tianjic/Tianmou co-design ecosystem and large-scale verification platform were upgraded, with industrial cooperation delivering controllable hardware for edge robotics, autonomous driving and AGI.