Research Associate · NIST Boulder
I build single-photon cameras, cryogenic readout electronics, and methods for training neural networks on physical hardware at NIST Boulder and the University of Colorado.
Superconducting nanowire single-photon detectors combine high sensitivity with low noise. Scaling them into cameras requires efficient readout as well as improvements in fabrication, detection efficiency, and count rate. Our camera combines thermally coupled row and column detectors with shared superconducting buses; pulse arrival-time differences identify the photon’s position.
Training a physical neural network can be difficult when its response differs from a differentiable model. Multiplexed gradient descent estimates how parameter perturbations affect a measured loss, allowing the network to learn from its own response without a backward pass through a hardware model.
Doctoral work on why superconducting accelerator cavities quench below their theoretical limit — a question about the first hundred nanometres of a niobium surface.
13 peer-reviewed journal articles in total. See the full list →
Demonstrates 400,000 imaging pixels, a 400-fold increase over previous SNSPD arrays. Four superconducting buses and eight microwave lines recover photon position from pulse arrival times.
Nature 622 (7984), 730–734 (2023) · 172 citations
Studies how multiplexed gradient descent scales with network size. In the tasks studied, time to reach a target accuracy need not increase linearly with parameter count and can decrease for larger networks.
APL Machine Learning 3 (2), 026107 (2025) · 5 citations
Pushes single-photon detectors to strips 0.1 mm wide — a thousand times wider than conventional nanowires — while keeping intrinsic detection efficiency, reshaping what geometries large-area detectors can use.
Optica 13 (8), 1649–1657 (2026)
25 pieces of coverage across two results, from NIST’s own newsroom to the photography and quantum trade press.
The big advance here is that the detectors are truly independent, so if you want a camera with more pixels, you just add more detectors to the chip.
In 2023 I led the demonstration of a superconducting nanowire camera with 400,000 pixels — roughly 400 times as many pixels as previous SNSPD arrays. The result came from abandoning per-pixel wiring: photon position is recovered from the arrival times of counter-propagating pulses on a shared readout line, an approach that keeps working as arrays grow. The work was published in Nature and covered widely in the science press.
Before NIST I completed a Ph.D. at the University of Maryland with Steven Anlage, building a near-field magnetic microwave microscope to find defects on superconducting RF cavity surfaces and running time-dependent Ginzburg–Landau simulations of how magnetic vortices nucleate there. I grew up in Tajikistan and studied physics at Boğaziçi University in Istanbul.
I am always glad to talk about superconducting detectors, cryogenic readout, or scaling hardware that was never meant to scale. Email is the surest way to reach me.
Applied Physics Division · NIST
325 Broadway, Boulder, CO 80305