Research & Development Areas

My research approach examines how computing systems should evolve as artificial intelligence, quantum technologies, and increasingly adversarial environments change the assumptions on which security, privacy, and trust have traditionally depended.

I tackle these challenges as end-to-end systems problems spanning cryptography, algorithms, software, computer architecture, and hardware. The emphasis is not only on developing new mechanisms, but on understanding how they move from standards and research prototypes into modular, interoperable, measurable, resilient, and deployable systems.

The current research program develops along four closely connected directions: post-quantum cryptography, privacy-enhancing technologies, confidential intelligence, and secure AI and hardware security. Across these areas, the broader objective is to make security, privacy, and resilience foundational properties of emerging computing infrastructure rather than capabilities added after systems have already been designed.

Post-Quantum Cryptography

The transition to post-quantum cryptography is a systems transformation rather than a simple replacement of cryptographic primitives. New standards must ultimately coexist with protocols, software interfaces, hardware, legacy infrastructure, and long-lived applications. Migration therefore raises questions of cryptographic agility, interoperability, assurance, performance, resilience, and maintainability across the computing stack.

This work studies the path from standards to deployment. The goal is to understand how quantum-resistant mechanisms can become dependable components of real systems while preserving explicit security boundaries, interoperability, reproducibility, and the ability to evolve as standards and deployment requirements change.

Active Open-Source Research Platforms (MIT License) include:

Together, these efforts treat post-quantum migration as more than algorithm adoption. They connect standards, software architecture, protocol composition, experimental evaluation, and resilient execution into a common systems problem.

Privacy-Enhancing Technologies

Privacy-enhancing technologies seek to reduce the exposure of sensitive information throughout computation, not only while data is stored or transmitted. My work spans encrypted computation, secure computation, trusted execution, privacy-preserving artificial intelligence, software systems, specialized architectures, and hardware acceleration.

A central focus is Fully Homomorphic Encryption, where computation is performed directly on encrypted representations. From a systems perspective, FHE changes encryption from a protective boundary into a computational substrate, creating new requirements for algorithms, compilers, memory systems, architecture, and hardware acceleration.

This research has explored the path from cryptographic capability to practical encrypted-computing infrastructure through programmable acceleration, processing-in-memory architectures, compiler and software systems, application integration, and international standardization.

Standards Development and Deployment

Practical adoption also requires common terminology, security expectations, interoperable representations, and implementation guidance. I therefore connect systems research with international standardization, including work on the ISO/IEC 28033 series for Fully Homomorphic Encryption and broader efforts toward secure and interoperable encrypted-computing systems.

Active Open-Source Research Platforms (MIT License) To Come Soon ...

Confidential Intelligence

Confidential Intelligence asks a complementary question: instead of continually adapting conventional algorithms to expensive encrypted execution, can we design computational paradigms that are inherently better matched to privacy-preserving computation?

Many algorithms were developed under the assumption that plaintext arithmetic, memory access, branching, precision, and communication are inexpensive. Those assumptions change substantially when computation takes place over protected or encrypted representations.

The research therefore explores algorithms and data representations whose computational structure aligns naturally with confidential execution. The goal is to co-design algorithms, cryptography, software, compilers, runtime systems, architectures, and hardware so that useful privacy-preserving applications can operate efficiently at scale.

A central direction is the exploration of computational structures that naturally map to encrypted arithmetic. Rather than treating the cryptographic substrate as a fixed backend, the application algorithms themselves become part of the design space.

Properties such as shallow computation graphs, regular data movement, associative operations, SIMD-friendly representations, and predictable communication may become important because they reduce the mismatch between conventional algorithms and encrypted execution.

The broader objective is not simply to demonstrate that an individual algorithm can execute under encryption, but to identify computational paradigms capable of enabling meaningful confidential applications at practical scale.

Active Open-Source Research Platforms (MIT License) To Come Soon ...

Secure AI and Hardware Security

Artificial intelligence increasingly relies on specialized hardware, accelerators, complex memory hierarchies, aggressive power management, and heterogeneous computing platforms. These mechanisms deliver substantial performance and efficiency gains, but they can also create physical and architectural attack surfaces that are invisible at the model or software layer.

Side-Channel-Resistant AI Hardware (joint work with NCSU)

One direction investigates how AI hardware can be designed to resist physical side-channel attacks. We study how information leakage emerges across algorithms, accelerator architectures, and physical implementations, and how protection mechanisms can be incorporated into the design of AI hardware itself.

The broader objective is to develop AI accelerators whose security properties are considered alongside performance and efficiency from the outset, rather than relying exclusively on protections added after the hardware has been designed.

Microarchitectural Physical Attacks on AI (joint work with NCSU)

A complementary direction investigates how microarchitectural mechanisms can create new physical attack surfaces against AI computation. Features introduced for performance and energy efficiency can produce observable behavior that reveals sensitive properties of models, workloads, and execution.

This work examines the interaction between AI workloads and the underlying processor microarchitecture to identify these emerging attack surfaces and understand how they can be mitigated without sacrificing the efficiency benefits that created them.

Faults and Resilient Encrypted Computation (joint work with TU Dresden and Barkhausen Institut)

Resilience is another dimension of trustworthy computing. Systems must preserve meaningful guarantees when computation is affected by natural faults, unreliable execution, or deliberate adversarial perturbations.

A recurring direction is the use of structure-aware detection and recovery mechanisms that exploit properties of the underlying computation rather than relying exclusively on generic redundancy. This work currently intersects strongly with post-quantum cryptography and encrypted computation, but the broader objective applies across secure and intelligent computing systems.

Active Open-Source Research Platforms (MIT License)

Systems Perspective

These research directions are deliberately interconnected. Post-quantum migration requires agility, interoperability, and resilient execution. Privacy-enhancing technologies depend on algorithms, software, architecture, and specialized hardware. Confidential Intelligence asks whether the algorithms themselves should change. Secure AI demonstrates why guarantees at one abstraction layer can fail when the underlying computing substrate remains exposed.

The common methodology is full-stack and experimental: build systems, expose assumptions, measure tradeoffs, validate interoperability and security, and use those results to inform standards, architecture, and deployment.

Much of this work is developed through the Confidential Intelligence Lab (CIL) at UC Irvine.