Silicon · Systems · Agents · AI for Hardware

Yangbo Wei 韦杨波

I am a second-year Ph.D. student in the School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University (SJTU), and the School of Computer Science, Eastern Institute of Technology, supervised by Prof. Lei He (IEEE Fellow). Prior to my journey in SJTU, I obtained my B.Sc. degree from Southeast University (SEU) in 2024.

I close the loop between AI systems and AI agents: better computing enables stronger intelligence, and stronger intelligence helps design better computing.

01 AI Agents & Reasoning

Agentic reasoning, Multi-Agent Systems (MAS), Efficient LLMs, self-evolving agents, agentic RL, agent infrastructure.

02 AI Hardware Acceleration

Energy-efficient AI inference, ASIC & FPGA design for AI, memory- and kernel-level optimization, heterogeneous acceleration, reconfigurable computing.

03 LLM System Optimization

Efficient LLM serving, parallelism & memory optimization, KV-cache management, speculative decoding, runtime scheduling.

04 AI for EDA

LLM agents for EDA, EDA foundation models, multi-fidelity optimization, AI4Science, and design automation intelligence.

Research Narrative · 2024–Present

From AI Chips to Self-Improving Agents

My research follows one connecting question: how can better computing systems enable more capable agents, and how can those agents, in turn, design better computing systems? I work across the full stack—from accelerator architecture and LLM serving to self-improving agents and AI-driven hardware design.

The AI-hardware flywheel: silicon accelerates computing systems, systems power intelligent agents, and agents automate the design of better hardware.
Faster silicon powers scalable systems and self-improving agents; those agents then automate the design of better hardware.
012024–

Silicon

I began with LLM acceleration on FPGA, ASIC, and heterogeneous architectures: expert-parallel MoE (ICCAD ’25), diffusion LLMs (ASPDAC ’26), HBM-aware multi-die systems (FPT ’25), and FPGA–GPU speculative decoding (ASPLOS ’26).

022026–

Systems

I now optimize inference across heterogeneous environments, from mixed-vendor GPU serving with Biren and on-device, multi-vehicle inference with Geely to semantic-structured KV-cache management for long-context reasoning (ICCAD ’26).

032025–

Agents

I build self-improving agent systems that evolve their own workflows, skills, tools, and reasoning strategies, including EvoMAS (ICML ’26), SkillSmith, and Mixture-of-Trees (AAAI ’26).

04Closing the loop

AI for Hardware

I apply these agentic capabilities back to hardware design through schematic synthesis (PCBgen, DAC ’26; PCBclaw, ICCAD ’26), Verilog generation (ASPDAC ’26), 3D-IC thermal simulation (DAC ’26), and the SmorFlux autonomous EDA agent framework.

Long-term goal. Build a self-reinforcing foundation where advances in systems and agents compound: better infrastructure produces stronger intelligence, and stronger intelligence creates better infrastructure.

🔥 News

  • 2026.04:  🎉🎉 Two papers were accepted by ICML 2026.
  • 2025.11:  🚀🚀 Two papers were accepted by ASPLOS 2026 (10.6%) and AAAI 2026 (17.6%).
  • 2024.05:  🏆🏆 Our paper received the Best Paper Award at ISEDA 2024!

📝 Publications

CCF-A Top Tier CCF-B Second Tier CCF-C Third Tier Award Award
17first / co-first papers
7CCF-A papers
7CCF-B papers

My work appears in top-tier AI, systems, and EDA venues including ICML, ASPLOS, DAC, AAAI, ACL, and ICCAD, spanning agentic reasoning, LLM acceleration, computer architecture, and AI for chip design.

Agentic AI & Reasoning

ICML 2026

Topological Active Inference for Task Disambiguation

Yangbo Wei, et al.

CCF-AML Top-3Active InferenceTask DisambiguationAgentic RL
ICML 2026

EvoMAS: Heuristics in the Loop--Evolving Smarter Agentic Workflows

Yangbo Wei, et al.

CCF-AML Top-3Multi-Agent SystemsWorkflow EvolutionSelf-Evolving Agents
ACL 2025

MECoT: Markov Emotional Chain-of-Thought for Personality-Consistent Role-Playing

Yangbo Wei, et al.

CCF-ANLP FlagshipRole-Playing AgentsEmotional CoTPersonality Consistency

LLM Systems & Architecture

ICCAD 2026

Cluster-of-Thought: Semantic-Structured KV-Cache Management for Long-Context Reasoning Models

Yangbo Wei, et al.

CCF-BEDA Big-4KV CacheLong-Context ReasoningStructured Semantics
AAAI 2026

Mixture-of-Trees: Learning to Select and Weigh Reasoning Paths for Efficient LLM Inference

Yangbo Wei, et al.

CCF-AAI FlagshipReasoning PathsEfficient InferencePath Selection
ASPLOS 2026

DFVG: A Heterogeneous Architecture for Speculative Decoding with Draft-on-FPGA and Verify-on-GPU

Yangbo Wei (Co-first), et al.

CCF-ASystems Top-tierSpeculative DecodingFPGA-GPUHeterogeneous Architecture
ASPDAC 2026

dLLM-OPU: An FPGA Overlay Processor for Accelerated Diffusion Large Language Models

Yangbo Wei, et al.

CCF-CAsia-Pacific EDADiffusion LLMOverlay ProcessorFPGA
ICCAD 2025

MoE-OPU: An FPGA Overlay Processor Leveraging Expert Parallelism for MoE-Based LLM

Yangbo Wei (Co-first), et al.

CCF-BEDA Big-4MoEExpert ParallelismFPGA Overlay
FPT 2025

FlightOPU: An FPGA Overlay Processor for LLM with HBM-Aware Multi-Die Architecture

Authors, ..., Yangbo Wei (3rd Author).

CCF-CReconfigurable ComputingHBMMulti-DieLLM Acceleration

AI for EDA

ICCAD 2026

PCBclaw: Hierarchical Multi-Agent Schematic Synthesis and Verification for MCU System Boards

Yangbo Wei (Co-first), et al.

CCF-BEDA Big-4Multi-Agent EDASchematic SynthesisMCU Boards
DAC 2026

PCBgen: Self-Evolving LLM Agent for Verifiable Specification-to-Schematic PCB Design

Yangbo Wei (Co-first), et al.

CCF-AEDA Big-4PCB DesignSelf-Evolving AgentVerification
DAC 2026

From Fluid Dynamics to Chip Design: Foundation Neural Operators Address Data Bottleneck in 3D-ICs Thermal Simulation

Yangbo Wei (Co-first), et al.

CCF-AEDA Big-43D-IC ThermalNeural OperatorsFoundation Models
ASPDAC 2026

Vflow: Discovering Optimal Agentic Workflows for Verilog Generation

Yangbo Wei, et al.

CCF-CAsia-Pacific EDAVerilog GenerationWorkflow SearchAgentic EDA
TODAES

ModelGen: Automating Semiconductor Parameter Extraction with Large Language Model Agents

Yangbo Wei, et al.

CCF-BACM JournalParameter ExtractionLLM AgentsSemiconductor Modeling
ISEDA 2024

Knowledge Transfer for GaN HEMTs Parameter Extraction Based on Hybrid Model

Yangbo Wei, et al.

Best PaperEDA SymposiumGaN HEMTsParameter ExtractionKnowledge Transfer

AI for Science & Healthcare

AAAI Oral 2026

MedAtlas: Evaluating LLMs for Multi-Round, Multi-Task Medical Reasoning Across Diverse Imaging Modalities and Clinical Text

Authors, ..., Yangbo Wei (3rd Author).

CCF-AAI Flagship OralMedical ReasoningMultimodal EvaluationClinical Text
ICASSP 2026

MambaDATG: Domain-Adaptive Tri-Plane-Gated Pre-Training for 3D Abdominal Segmentation

Yangbo Wei (Co-first), et al.

CCF-BSignal Processing Flagship3D SegmentationDomain AdaptationTri-Plane Gating
BIBM 2025

H3DE-Net: Efficient and Accurate 3D Landmark Detection in Medical Imaging

Yangbo Wei (Co-first), et al.

CCF-BBioinformatics VenueLandmark DetectionMedical ImagingEfficient 3D

🎖 Honors and Awards

Academic Awards

  • 2026 Travel Grant Award, ICML 2026 (International Conference on Machine Learning)
  • 2026 Silver Reviewer Award, ICML 2026 (International Conference on Machine Learning)
  • 2024.05 Best Paper Award, ISEDA 2024 (International Symposium on Electronic Design Automation)

Competitions

  • 2023 First Prize (Meritorious Winner), Mathematical Contest in Modeling (MCM/ICM)
  • 2023 Second Prize, China Undergraduate Mathematical Contest in Modeling (National Level)
  • 2022 First Prize, Southeast University Electronic Design Competition

💼 Experience

Academic Service

  • 2026 Reviewer, ICML 2026 (International Conference on Machine Learning)
  • 2026 Reviewer, NeurIPS 2026 (Conference on Neural Information Processing Systems)

📖 Education

  • 2024.09 - Present, Ph.D. Student in Integrated Circuit Engineering, Shanghai Jiao Tong University (SJTU).
  • 2020.09 - 2024.06, B.Sc. Southeast University (SEU).

🏖️ Life & Misc

“Don’t Panic. In a universe this big, a bug is nothing.”