Tanvir M. Mahim

Tanvir Mahmud Mahim

Adjunct Lecturer and Research Assistant, Department of Electrical and Electronic Engineering, BRAC University. I build physics-based computer models of quantum and semiconductor devices that do not stop at the device: they follow the signal through the instrument, all the way to the number a measurement actually returns.

Each device I study is described once, from the material it is made of to the signal its measurement produces. A model built this way works in both directions. Run forward, it acts as a referee: when two experiments on the same sample disagree, it can tell how much of the difference comes from the material and how much from the measurement itself. Run backward, it designs: the model can report how its output responds to every input, so an optimizer can turn a target specification directly into a device, and can tune the measurement setup too, not just the device. This approach has carried my work from 2D and GaN electronics to quantum photonics. It is now aimed at the step that limits quantum hardware most: the measurement itself.

Squeezed-light microcombs, diamond quantum sensing and photonic machine learning, conducted with Dr. A. S. M. Mohsin. Squeezed light is laser light whose noise has been pushed below the usual quantum limit; it is a raw material for precision measurement. Our first published result in this area appeared in Optics Express (2026), and a second, FabGAN-ID, in the IEEE Sensors Journal (2026). FabGAN-ID learns how a deposition process really errs, from its own records, and designs optical filters that survive those errors. A single micro-ring on a silicon-carbide chip can make such light, but building the light up inside the ring and letting it out work against each other, so one ring can never deliver more than 3 dB. Adding a second ring, whose only job is extraction, releases 7.9 dB. The other project listens to magnetic noise with a diamond sensor. Vortices, tiny whirlpools of magnetic flux, waste energy in the tantalum films used in the best superconducting qubits. But not all trapped flux is lossy. A basic law of physics says that a vortex that wastes energy must also jitter, while flux held in a patterned hole stays silent. Our calculations show that a diamond spin sensor 50 nm above the film would hear one moving vortex clearly, about 6,000 times above the film’s own noise. The noise then gives the loss of each vortex, so the lossy ones could be counted before a qubit is cooled. A third project, SPARQ, published in ACS Photonics (2026), speeds up the search for single-photon light sources. In simulation, a learning method that reads the data as they arrive reaches 90% accuracy with six times less measuring time than a standard fit. It can often decide a third of a second into a one-second measurement, at 86% accuracy. A fourth project designs a fast readout for a tin-vacancy spin in diamond, a quantum memory that talks to light. In simulation, a deliberately spoiled nanocavity on a lithium niobate chip reads the spin in 0.1 µs with 98.8% fidelity, against a measured 87.4% in 50 µs with a microscope.

GaN and 2D semiconductor devices and physics-guided design, conducted with Prof. Md. Mosaddequr Rahman. Electric charge and heat-carrying vibrations are followed through nitride and single-atom-thick channels, starting from the underlying quantum states rather than from shortcuts. Three findings stand out. In the GaN/AlN system, published measurements disagree on the masses of the holes. In a paper now accepted in Physical Review B, our calculation matches the heavy-hole mass with nothing adjusted, and shows why one experiment reads it higher: there the holes scatter before they can finish an orbit. It also shows that only one band parameter can explain the light-hole puzzle, and it identifies the kind of disorder the holes scatter from. In patterned single layers, two Raman peaks show that a cut MoS2 edge holds extra electrons but no measurable strain. Published transistor thresholds then show that the fixed charge there is negative and reaches about 12 nm into the ribbon. In a model of a thin-gate transistor, using this charge instead of the Raman value raises the predicted width at which the current per width halves from 17 to 96 nm, mostly through a threshold shift that gate design can offset. And for a WSe2 ferroelectric memory transistor, in a paper now published in Materials Today Communications, we show when the channel’s own properties drop out of its memory window. In the model, one percent of compression then more than doubles how easily charge moves, while the 1.43 V memory window moves by less than 0.1 percent. Strain and memory can therefore be tuned separately; on the channel side, charge trapped at the surface is what limits the memory.

Within this area, a collaboration with Dr. Nadim Chowdhury at BUET carries the approach into physics-guided circuit design. Here the design software computes how every circuit output responds to every design choice, so an optimizer can adjust a whole circuit automatically. Two examples. A single-chip GaN frequency synthesizer (the circuit that generates a radio’s precise clock signal) stays locked from −55 to 150 °C in transistor-level simulation, and designing it revealed three device rules that set the temperature limits of such chips. And a machine-learning framework built with GlobalFoundries, Inc. sizes the transistors of analog circuits, and carries what it has learned across four different manufacturing technologies. Alongside this I worked as a consultant to GIXSystems.ai on a 100 V GaN power transistor, fabricated with X-FAB in Erfurt.

Learning-based grid control and bifacial photovoltaics, conducted with Dr. A. H. M. A. Rahim, who retired in December 2024. Power-grid controllers that revise their own decision rules as they learn hold a small local grid and a wind turbine steady through faults severe enough to defeat conventional fixed-setting controllers. The same collaboration produced a custom-built two-sided solar module and the efficiency model that describes it. This is the earliest of the three research areas, and the one with the longest published record.

I completed the B.Sc. in Electrical and Electronic Engineering at BRAC University in December 2022 with High Distinction, and joined the department as a Teaching Assistant, then as a Research Assistant, and since June 2025 as an Adjunct Lecturer. The Research page presents all eighteen projects in full, with figures, code and data.

I welcome correspondence on any of these topics, as well as inquiries regarding collaboration, at tanvir.mahim@bracu.ac.bd.

Research interests

  • Quantum Optics and Photonics
  • 2D Quantum Materials
  • Wide Bandgap (WBG) Semiconductor Devices
  • Quantum Physics
  • Electronic Design Automation (EDA)
  • Machine Learning

News

  • Oct 2026 Our paper "Origin of the conflicting hole masses in the GaN/AlN two-dimensional hole gas" is accepted in Physical Review B. Three kinds of measurement disagree on the masses of the holes trapped at a GaN/AlN interface. Our calculation gives a heavy-hole mass of 1.92 to 1.99 m0, matching the measured 1.92 ± 0.16 m0 with nothing adjusted. It also shows why the cyclotron-resonance value differs: that resonance is overdamped. Paper · Details
  • Oct 2026 Our paper "Closed-Loop, Event-Driven Machine Learning for Autonomous Triage of Single-Photon Emitters", with the University of Memphis, is published online in ACS Photonics. It asks how much measuring time can be saved when screening candidate single-photon emitters. In simulation, a network that reads the data as they arrive reaches 90% accuracy with 6 times less measuring time than a standard fit, and on published quantum-dot data a network trained only on simulated data halves the error of the usual analysis. Paper · Details
  • Oct 2026 Our paper "Channel decoupling of the memory window in CuInP2S6-gated two-dimensional ferroelectric transistors: conditions, limits, and implications for interface processing" is now published in Materials Today Communications (volume 57, article 116333) and is open access. It shows when the memory window of a ferroelectric transistor does not depend on its channel, and how strain and interface traps act. Paper · Details
  • Oct 2026 Our paper "FabGAN-ID: Learning the Fabrication Process for Yield-Aware Inverse Design of Multilayer Photonic Sensor Filters" is published online (early access) in the IEEE Sensors Journal. A generative model learns how a deposition process errs from 400 process traces, and the filter is then designed against it. In this fully simulated study, the pass/fail yield of a 532 nm notch filter rises from 0.16% to 63.7%. Paper · Details
  • Sep 2026 All twelve open-source tools have new releases. Most fix results that could be wrong without a warning, and several add tools for working with measured lab data; each release lists its changes with before-and-after numbers, and every new result is checked by public tests. Details
  • Sep 2026 The documentation of all twelve open-source tools has been rewritten in plain language, with worked examples whose printed output is checked by running them. Checking every document line by line against its code also found real bugs; each is fixed, with a new test, in a patch release of every tool, and the release notes list them. Details
  • Sep 2026 Two new tools join the open-source family, now twelve: fracpll (design a fractional-N frequency synthesizer from your own measured tuning curves and noise numbers, with the loop model refusing outside its own validity) and lockkernel (exact thresholds and exponents of synchronization transitions, plus fits of measured synchronization curves with honest error bars). fracpll grows out of the GaN-on-SOI PLL co-design study on the Research page, and lockkernel is the maintained release of the reference code behind the locking-kernel universality study; both follow the same rule as the rest: their main results are checked by public tests against closed forms or exact simulations. Details
  • Sep 2026 The open-source family is now ten tools: vacspin (diamond colour-centre spin-photon interfaces, with cited parameters for the silicon-, germanium- and tin-vacancy centres) and labplan (plan, fit and certify a calibration of any user-supplied model) join the eight released earlier. Every package now carries a laboratory layer that plans measurements before they are taken and calibrates from the lab's own data with honest uncertainties. Details
  • Sep 2026 A complete written course now covers all eight open-source tools: Research-Grade Computational Physics in Python, a 66-page book with 25 runnable code lessons, one module per tool. Every number printed in the book was produced by running the included scripts against the released packages, and each script checks its own textbook results. Available in the Patreon shop.
  • Sep 2026 The open-source family is now eight tools: hamop (one material model, every property) and fabtwin (learning a fabrication line's real errors to design for manufacturing yield) join the six released earlier, and all eight shipped major feature releases this month. Details
  • Sep 2026 Our paper "Overcoming the 3 dB squeezing extraction limit in silicon carbide microcombs with a photonic molecule" is published in Optics Express (open access). A second ring lets a microcomb release 7.9 dB of squeezed light, where a single ring can never exceed 3 dB. Paper · Details
  • Sep 2026 The open-source family grows to six tools: cavsqueeze (spin squeezing of large ensembles in resonators) and SPARQ (automated screening of single-photon emitters) join ramansep, kpenvelope, sqzcomb and absnoise on PyPI, each tested in public and archived with a DOI. Details
  • Aug 2026 Received the Quality Journal Publication Award from the Research Metrics Committee of BRAC University for a second successive cycle. The award recognizes research published in Q1 journals, those ranked in the top quartile of their field.
  • Aug 2026 Released absnoise, a fourth open-source tool, covering the occupation noise of Andreev bound states and the sensitivity limits of the proximity Josephson thermal detectors built on them. All four tools are installable from PyPI and archived on Zenodo under versioned DOIs. Details
  • Aug 2026 All three research tools are now installable directly from PyPI (pip install ramansep, kpenvelope, sqzcomb) through automated, tested release pipelines. sqzcomb v0.2 adds the photonic molecule, with its test suite driving a coupled two-ring design past the single-ring 3 dB detected-squeezing limit. Details
  • Aug 2026 Released three open-source research tools under the TaN-MM-Org organization: ramansep, kpenvelope and sqzcomb, each with a tested core and continuous integration. Details
  • Jun 2026 Our paper on adaptive Takagi–Sugeno–Kang fuzzy inference with reinforcement learning for low-voltage ride-through of DFIG wind turbines is published in Engineering Science and Technology, an International Journal. Details
  • Dec 2025 Joined GIXSystems.ai (USA) as a consultant, working on the technology development of a 100 V GaN-on-Si power HEMT discrete device, with tape-out at X-FAB in Erfurt, Germany.
  • Late 2025 Received the Quality Journal Publication Award from the Research Metrics Committee of BRAC University, for research published in Q1 journals.
  • Sep 2025 Our review of the challenges and prospects in agrivoltaics appears online in Advanced Energy and Sustainability Research (Wiley, open access); it runs in the January 2026 issue.
  • Aug 2025 BRAC University featured our work in its news, highlighting the IEEE Transactions on Fuzzy Systems and IEEE Sensors Journal papers with Prof. Md. Mosaddequr Rahman and Prof. A. H. M. A. Rahim.
  • Jul 2025 Our hierarchical inverse-design framework for unit-cell CMUTs is published in the IEEE Sensors Journal, with the dataset released on Zenodo.
  • Jun 2025 Appointed Adjunct Lecturer in the Department of EEE, BRAC University, teaching Solid-State Devices, Numerical Methods, and Microprocessor & Embedded Systems.
  • May 2025 Started work on physics-guided circuit design with Dr. Nadim Chowdhury at BUET, including a collaboration with GlobalFoundries, Inc. (USA) on graph reinforcement learning for transferable analog transistor sizing.
  • Feb 2025 Our adaptive fuzzy attention inference controller for microgrids under extreme grid-bus faults appears in IEEE Transactions on Fuzzy Systems.
  • Jan 2025 Began research on squeezed-light microcombs, diamond quantum sensing and photonic machine learning with Dr. A. S. M. Mohsin, covering soliton-crystal microcombs, NV micromembrane vortex sensing, and closed-loop machine learning for quantum photonic instrumentation. Read more
  • Jul 2024 Began research on GaN and 2D semiconductor devices and physics-guided design with Prof. Md. Mosaddequr Rahman, spanning the GaN/AlN two-dimensional hole gas, nanoribbon transistors, ferroelectric WSe2 logic, and fabrication-aware inverse design.
  • Jul 2024 Our weather-responsive multidimensional efficiency model for custom-built bifacial panels is published in the IEEE Journal of Photovoltaics.
  • Mar 2024 Our comparative review of mono- and bifacial photovoltaic technologies appears in the IEEE Journal of Photovoltaics.
  • Aug 2023 Profiled by the Aspire Institute in its First-Gen Friday series, during the Aspire Leaders Program, which was founded at Harvard University in 2017 and became an independent institute in 2021.
  • Dec 2022 Completed the B.Sc. in Electrical and Electronic Engineering at BRAC University with High Distinction.