Yoshi Sato

I am an undergraduate researcher at Waseda University. My research focuses on building privacy-preserving and trustworthy systems, spanning TEEs, confidential computing, and cryptographic protocols. I am increasingly interested in microarchitectural security, recognizing it as the critical foundation on which these systems depend.

I am currently a visiting research intern with the Decentralized Systems Group at Yale CS, advised by Prof. Fan Zhang, where I am leading the technical implementation of a privacy-preserving bug bounty platform for smart contracts. I am also a research assistant at Internet Initiative Japan, advised by Dr. Pierre-Louis Aublin, working on serverless confidential containers in collaboration with a PhD student from Imperial College London’s Large-Scale Data & Systems Group.

In a world where data is the new currency, trust and privacy are no longer optional—they are foundational. I am driven by the question: how do we build infrastructures that are not only fast and scalable, but fundamentally trustworthy? My motivation is to advance secure systems from hardware through software, guided by the belief that trust must be a first-class property of modern computing.

News

Aug 16, 2025	My poster paper was accepted to IEEE DASC 2025!
July 24, 2025	Visiting New Haven, CT for my research internship!
June 20, 2025	Volunteering at ISCA 2025, held this year at Waseda University.

Publications

IEEE DASC

Towards GPU Passthrough in Intel TDX: Design Challenges and Early Baselines

Y. Sato, H. Uranami, A. Saiki, K. Kimura

The 23rd IEEE International Conference on Dependable, Autonomic and Secure Computing, Hokkaido, Japan.

Abs PDF

Confidential Virtual Machines (CVMs) like Intel TDX and AMD SEV-SNP provide hardware-based isolation for cloud workloads. While SEV-SNP supports GPU passthrough for legacy (non-CC capable) accelerators, Intel TDX currently lacks this capability. This work-in-progress analyzes the perfor- mance implications of enabling such functionality across both platforms. We identify fundamental architectural differences: SEV-SNP’s guest-controlled page encryption enables near-native performance, while TDX’s hypervisor-mediated design requires mandatory VMEXITs and may impose significantly higher costs. Our CUDA microbenchmarks establish baseline metrics showing pageable transfers achieve 50-75% throughput of pinned DMA on Intel Xeon systems. We present the technical requirements for TDX GPU passthrough and analyze why Intel’s stricter security model complicates performance. This work provides developers with critical performance expectations when choosing between CVM platforms for GPU workloads, highlighting the trade-offs between security architectures and practical performance.

Research Projects (ongoing)

Project

DeadDrop: Privacy-Preserving Bug Bounty Platform for Smart Contracts

Collaborators: Yoshi Sato, Prof. Fan Zhang (Yale), PhD students from Yale, Cornell, and other institutions.

Status: Ongoing project.
Role: Lead technical implementation.
Focus: Using OMR cryptographic primitives and TEEs to secure multi-party bug retrieval

Abs Code

Bug bounty programs are widely used to uncover software vulnerabilities, but they face unique challenges in the context of smart contracts, where pseudonymous authors and financial stakes complicate responsible disclosure. We present DeadDrop, a system for oblivious bug reporting that combines Oblivious Message Retrieval (OMR) with Trusted Execution Environments (TEEs). DeadDrop allows researchers to submit vulnerability reports privately and efficiently without exposing recipients or signaling the presence of bugs. Our design includes techniques for long messages and a bug specification language, and our prototype demonstrates that oblivious bug reporting for smart contracts is both practical and efficient.
Project

SC2: Serverless Confidential Containers

Collaborators: Yoshi Sato, Pierre-Louis Aublin (IIJ), Carlos Segarra (Imperial College London), and members of the LSDS Group.

Status: Ongoing project, presented in internal seminar.
Role: Research assistant and collaborator.
Focus: Enabling privacy-preserving serverless frameworks with knative, k8s, and confidential containers.

Abs Code Slides Video

Serverless computing offers elasticity and cost efficiency, while confidential VMs (CVMs) provide strong hardware-backed isolation. Combining the two promises privacy-preserving serverless platforms, but faces a fundamental challenge: CVMs incur high startup costs that break the performance model of serverless. We present SC2, a prototype serverless platform built with Knative, Kubernetes, and Confidential Containers. SC2 introduces a cache layer that enables efficient warmup and reuse of CVMs, mitigating cold-start latency while preserving confidentiality guarantees. Our early results demonstrate that caching can make confidential serverless practical, opening a path toward scalable privacy-preserving cloud services.