NeuroCore

AFE: Energy-harvesting power converter
AFE: Inductive sensing
Analog front-end (AFE)
ADC: Comparator Biasing
ADC: Register file
ADC: Transconductor-based DAC
ADC: Comparator
ADC: Demux for updated bit values
AFE: Custom ADC topology
ADC: Successive-approximation logic
ADC: Register unit (feat. disturbance-resistive SRAM cells)
ADC: full-range Spectre sims
Power converter: stress testing Spectre sims (upper voltage range)
Power converter: stress testing Spectre sims (lower voltage range)
Power converter: stress testing Spectre sims

Inspiration

Current TMS therapies are "open-loop" because doctors set a dial on the machine, but they do not actually know how the magnetic field behaves once it hits the physical tissue. To fix this, we need a way to measure the field from the inside. We envisioned NeuroCore: a battery-free, wireless ASIC that lives inside the cranium, harvesting energy from the TMS pulses to act as a real-time error controller, ensuring the tissue receives the exact magnetic dose intended.

What it does

Monitors the Field: It uses an ADC interface (adc_data[3:0]) to capture the raw magnetic field output directly at the tissue level, reading exactly how the tissue is absorbing and distorting the TMS pulse.

Analyzes Distortion: A Haar-Lifting DWT Engine decomposes this field data into 8 frequency subbands to identify the dominant distortion bin.

Acts as an Error Controller: Instead of just reporting what it sees, the chip compares the measured dominant distortion bin against a desired target setting (target_idx[2:0]).

Auto-Adjusts Therapy: Based on that comparison, it issues a proportional 3-bit correction command (cmd_out) back to the TMS machine via LSK backscatter, telling the machine exactly how to adjust its output to hit the target.

Stays Safe: It solves the thermal issue by processing all the error-correction math locally, only transmitting tiny, sparse 14-bit command packets instead of a continuous data stream.

How we built it

Proportional Control Logic: We designed a digital comparator system within the Main FSM that calculates the delta between the measured bin and the target index to generate the real-time correction command.

Efficient Math: We replaced heavy CORDIC math with a simple Absolute-Value Magnitude Bank and a time-shared single multiplier for power accumulation, saving critical chip area.

The Wireless Link: We implemented a Manchester-encoded LSK (Load Shift Keying) Modulator. By toggling the impedance of the internal coil, the chip backscatters the correction commands to the external TMS machine without a single wire.

Challenges we ran into

Interference vs. Measurement: We had to balance harvesting enough magnetic energy to power the chip logic while simultaneously using that same magnetic field as our primary measurement variable without saturating the ADC.

Power Consistency: Designing a synchronous error-controller that stays stable while being powered by the transient, pulsing magnetic waves of a TMS machine required precise timing and power-gating control.

Accomplishments that we're proud of

Closing the Delivery Loop: We successfully moved TMS from an open-loop guess to a closed-loop certainty, guaranteeing the magnetic field hitting the tissue perfectly matches the clinical target.

Battery-Free Operation: We conceptualized a design that minimizes components by removing the battery, making the implant significantly safer and smaller.

What we learned

Algorithm Hardware Co-Design: We found that simplifying complex math (like DWT) for hardware not only saves power but can actually make signal detection faster.

Multiplier-Free Math (Bit-Shifting): We learned how to drastically cut power and gate count by avoiding standard hardware multipliers. Instead, we used arithmetic right bit-shifts (>>>) to apply fractional coefficients (like 0.25 and 0.125), proving that deep optimization happens at the bit level .

Manchester Encoding: We learned how to write a digital modulator from scratch to encode our 14-bit command packets . By making the first half of the bit period the data and the second half the inverted data, we ensured the wireless signal is DC-balanced and self-clocking .

Regulatory Complexity: Developing implantable tech in 2026 requires strict adherence to safety standards like IEC-60601 to prevent electromagnetic interference.

What's next for NeuroCore

Capturing True Neural E-Waves: Now that we have a closed-loop system that can perfectly control and measure the magnetic field distortion (Step 1), our next major milestone is upgrading our analog front-end (AFE) to capture the actual neural electrical signals (e-waves) reacting to that field (Step 2).

Advanced AFE Blanking: To read delicate microvolt e-waves right after a massive TMS magnetic pulse, we will design custom analog blanking circuitry. This will protect our amplifiers from saturating during the pulse so our digital filters can instantly start processing the brain's reaction.

Dual-Modality Feedback: Ultimately, we want to combine our magnetic error-correction data with true neural firing rates. This will give the closed-loop system a multi-dimensional view of both the physical tissue state and the neurological response.

Appendix: Analog Front-End (AFE) Overview

The AFE features a spiral on-chip inductor, which converts the TMS magnetic field into current, an analog-to-digital converter (ADC), and a power converter circuit that provides power to the entire system by harvesting sharp voltage spikes across the inductor.