Undergraduate at Nanjing University (Integrated Circuit Design and Integrated Systems), graduating in 2027.
I work on autonomous AI systems, AI accelerators, computer architecture, and hardware-software co-design.
Personal website: ahappend.github.io
- Nanjing University — Integrated Circuit Design and Integrated Systems
- Expected graduation: 2027
Autonomous AI DNN Mapping Scheduling NoC RTL Edge AI Reproducible Systems
- Native AWQ W4A16 and mixed-precision RTL for transformer inference
- DNN mapping and inter-layer scheduling for tiled accelerators
- Hardware-software co-design for AI accelerators
- Evidence-driven autonomous research and engineering systems
- Argus/original/Argus Agent/Microsoft - Open multi-agent runtime for long-horizon research and engineering, combining persistent state, reviewed execution, verification, and evidence. Here is the Technical Report.
- ACE-3/personal/ACE-3/organization (active development) - An evolving open research RTL project for native AWQ W4A16 transformer inference, with independently checked arithmetic, full-input projection, FP16 adaptation, QKV/RoPE/KV-cache, attention, and decoder building blocks. This is a research preview, not yet a complete 24-layer accelerator or hardware-performance result.
- ACE-2/personal/ACE-2/organization - Argus-built Qwen2.5-0.5B W4A8 accelerator with SystemVerilog RTL, certified two-token integration, SKY130 100 MHz results, and evidence-first host-trust recovery.
- 64-Point FFT - Resource-efficient radix-2 DIT FFT in SystemVerilog using folded streaming SDF RTL, Q1.15 fixed-point arithmetic, Vivado/XSim simulation, Python/NumPy co-verification, and Virtex-7 implementation results.
- Two-Stage CMOS Op-Amp - Two-stage CMOS operational amplifier in TSMC 0.18 um technology, with Miller compensation, Spectre AC/DC/transient verification, and analysis of gain, bandwidth, phase margin, slew rate, output swing, and power.
- HyperCut: Fast Inter-Layer Scheduling via Directed Hypergraph and Early Filtering — Academic paper on early filtering and hypergraph-based inter-layer scheduling for tiled DNN accelerators.
arXiv: https://arxiv.org/abs/2608.19296 - Argus: A General-Purpose Agentic Reasoning Runtime for Long-Horizon Tasks — Technical report on a persistent, self-evolving runtime for long-horizon agentic reasoning.
arXiv: https://arxiv.org/abs/2608.05144
Python C/C++ TypeScript SystemVerilog Verilog React Linux
Feel free to contact me!
- Email/school(Nanjing University)
sufeng_guo@smail.nju.edu.cn - Email/gmail
hapouterwall@gmail.com - Email/qq
sufeng_guo@qq.com