Shashank Bansal

    PhD candidate, Bioengineering · UC San Diego

    Shashank Bansal

    शशांक बंसल

    I build the sensors and silicon that listen to the brain: ultra-low-power, high-density recording chips that turn neural activity into action.

    Scroll to begin · drag the brain to rotate it

    01Sensing

    Two ways to listen

    Every brain–computer interface starts with a sensor. Outside the skull you trade resolution for safety; inside, you can hear single neurons.

    Non-invasive

    Invasive

    ★ The four I work with: EEG, MEG, HD-DOT and intracortical

    02Signals

    Microvolts, thousands of times a second

    Each electrode carries field potentials, high-gamma bursts and spikes, buried in noise at tens of microvolts. Every channel has to be amplified, filtered and digitized right next to the tissue, in parallel.

    12 channels shownrecording
    100 µVsynthetic signal

    03The chip I build

    Up to a thousand channels, a microwatt each

    My neural interface system-on-chip puts a low-noise front-end and digitizer behind every electrode, with impedance monitoring built in. I'm integrating it with transparent Neuro-clear probes for simultaneous electrophysiology and optical imaging. Hover the die to explore it.

    up to 0channels
    0µW per channel
    0noise efficiency factor
    0mW at 1,024 channels

    05Decode, act, feed back

    Closing the loop

    Hundreds of channels stream to the chip in parallel and are decoded into intent. Here, a person with paralysis moves a cursor across their communication board and selects what they need, and every selection sends a stimulation pulse back to the cortex as feedback. Try being the decoder ↓

    Neural features channel × band
    Decoder intent
    Communication board stimulation

    ▍

    06Your turn

    Be the decoder

    Closing the loop means a real person doing a real task. Here, someone with paralysis wants to ask for something on their communication board. Three neurons on one electrode carry their intent: one fires for left, one for right, one for select. Sort each spike while it is in the sort window with ← → ↑ (or tap the buttons), and leave artifacts alone. Mislabel a spike and the cursor goes the wrong way.

    They want: WaterRequests done: 0

    Three neurons, one electrode, one person

    Match each spike's shape to a template on the right. Every correct sort moves their cursor; selecting the item they want completes a request. 60 seconds, and every round deals new neurons.

    ← unit 1   → unit 2   ↑ unit 3

    ⏱ 500× slower than real time. One second of play is two milliseconds of recording (see the axis). A real spike lasts about 1.6 ms, far too quick to sort by eye, so the game slows the recording down. Spikes are also packed closer together than a real neuron would fire.

    Portrait of Shashank Bansal in an Illinois cap and sunglasses, on a Catalina cliff above the ocean

    About

    Engineering the interface between brains and machines

    I'm a first-generation PhD candidate in Bioengineering at UC San Diego, advised by Prof. Gert Cauwenberghs in the Integrated Systems Neuroengineering Lab. I work on dynamic interfaces between living and artificial nervous systems.

    The brain is one of the most complex systems we know of, and I'm trying to science and engineer the heck out of it. My focus is the hardware layer: the sensors, the low-power mixed-signal chips that record from them, and the in-memory and neuromorphic circuits that make sense of the data.

    Outside the lab I hike, ski, play squash and take photos.

      Research

      What I work on

      Selected publications All on Google Scholar ↗

        Patents

          Posters & talks

            Experience

            Path so far

              PDF ↓