I am an associate professor at Shanghai Jiao Tong University. I received my Ph.D. degree in computer science from Shanghai Jiao Tong University in 2020. My research area includes the agent OS, system for LLM and heterogeneous systems.
Agent OS
I build high-performance, secure, and reliable agent runtime frameworks for agent applications such as mobile agents and embodied agents. My research covers high-performance agent runtime frameworks, reliable agent decision-making mechanisms, and agent evaluation systems.
Inference System
I build high-performance and secure inference frameworks for large models such as Large Language Models (LLMs) and Diffusion Models.
RS-Infer[ATC'26]: builds a fast and secure LLM serving system for mobile devices with flexible resource isolation (Flex-Mem and Flex-NPU) within Arm TrustZone.
MosaicKV[arXiv'26]: serves long-context LLMs with dynamic two-dimensional KV cache compression. [
pdf ]
TZ-LLM[EuroSys'26]: protects on-device LLMs with Arm TrustZone via parsimed CMA allocation and a TEE–REE co-driver for secure NPU sharing. [
pdf ]
Ongoing work on super-node LLM system, dLLM optimization, quantization, and operator fusion for large-model inference.
HW/SW Co-Design for System Security
I use hardware-software co-design to improve the security and performance of system software, including operating systems, virtualization systems, and system services. By introducing architectural support and co-designing it with the software stack, I build more efficient isolation and trusted execution frameworks.
HSMEM[MICRO'26]: provides scalable and energy-efficient data transfer for heterogeneous confidential memory between CPU and GPU.
EIDS[TOCS'26]: builds a high-performance and maintainable cloud intrusion detection system. [
pdf ]
XpuTEE[TOCS'25]: builds high-performance and practical TEEs across CPU and GPU for heterogeneous computing.
ISA-Grid[ISCA'23]: constructs bit-level isolation domains for instructions and registers to isolate different kernel components and kernel modules. [
pdf ]
CPS[ASPLOS'23]: makes VM isolation more efficient in many-core systems via cooperative para-virtualized scheduling. [
pdf ]
COLONY[TC'22]: constructs TEE instances at the kernel level to protect
OS services. [
pdf ]
TZ-Container[SCIS'21]: isolates containers from an untrusted
OS with ARM TrustZone. [
pdf ]
XPC[ISCA'19][TOCS'22]: provides architectural support for secure and efficient cross-domain calls to speed up IPC for existing OSes. [
pdf ] [
pdf ]
EPTI[ATC'18]: isolates the kernel of unpatched VMs from user-level Meltdown attacks using the Intel VMFUNC instruction. [
pdf ]
vTZ[Security'2017]: the first secure virtualization system for ARM TrustZone, enabling TrustZone for each VM. [
pdf ]
SGXMigration[DSN'2017]: securely migrates a VM together with its TEEs on untrusted clouds. [
pdf ]