2025 — now
Northwestern
finding
Side channels on quantum decoders
Fault-tolerant quantum computing depends on a classical decoder that
interprets error syndromes in real time. I showed, for the first time on
real hardware, that this decoder's execution time and power draw leak the
syndrome it is processing — the error-correction machinery betrays the
errors it exists to hide.
To measure it I ported the distributed union-find decoder (Helios) and a
constant-latency lookup-table decoder to a Xilinx Artix-7 and built the
full measurement pipeline, capturing per-decode timing and power across
hundreds of thousands of decodes of real Google Willow surface-code
syndromes at distances 3 and 5. From timing alone the traces fingerprint
individual IBM Heron processors with up to 89% accuracy, and recover code
distance, logical error rate, and detector-firing structure.
2024 — now
Northwestern
system
CAPIO — giving applications the device, not the keys to it
Kernel bypass lets an application talk straight to a NIC or NVMe drive,
which is fast, but the registers it legitimately needs often share a page
with registers it absolutely should not touch. CAPIO is a capability-based
design that grants userspace exactly the MMIO registers it is entitled to
and isolates the privileged ones at the same granularity, on commodity
hardware.
I built the kernel and device-driver mechanisms underneath it, and mentor
the students working on the systems implementation.
arXiv:2512.16957
ROSS @ SC26, to appear
2024 — now
Northwestern
system
TrustCheckpoints — a root of trust with no secret to steal
Attestation normally rests on a key burned into the device or a dedicated
trusted element, both of which are things an attacker can extract or a
manufacturer can get wrong. TrustCheckpoints needs neither: a verifier
issues randomized polynomial memory challenges whose timing a compromised
device cannot fake, so malware gives itself away by being too slow.
The prototype is end to end — firmware, hypervisor, Linux, and an external
verifier — running on ARM Cortex-A53.
2022 — 2025
Pitt · Northwestern · Michigan
system
Privacy Backplane — inverting DRM so people set the policy
Ubiquitous sensing is already a fact of life: large fleets of IoT devices
quietly record and analyse personal information about whoever walks past
them. Blanket legal frameworks like GDPR and CCPA answer this with one
policy for everyone. The Privacy Backplane is a legal-technical framework
for the opposite — environments must negotiate a clear policy with each
individual covering what may be collected about them and how it is used,
enforce it, and then certify that they did.
It is classic DRM with the roles reversed. The individual is the content
producer, and holds both the knowledge of and the control over the policies
attached to their data. I built systems mechanisms underneath it, including
an ARM/Linux video pipeline and the kernel-level components that enforce
policy on the device itself.
