A Closed-Loop Neuromodulation Chipset with 0.0009\text{mm}^{2}-0.36 \mu \mathrm{W}/\text{Ch}$ Recording Frontend and ${0.075} \mathrm{m m}^{\mathrm{2}}-\mathrm{6.76}{\mu}\mathrm{W}$ Seizure Classification Backend

This work presents a closed-loop neuromodulation chipset, integrating 64 analog frontends (AFEs) featuring a novel IDAC-embedded OTA, a lightweight feature extraction unit, a Sparse Projection Oblique Randomer Forest (SPORF) classifier, and 4 high-voltage (HV) compliant current stimulators. Each AFE occupies <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.0009 \text{mm}^{2}$</tex> and consumes <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.36 \mu \mathrm{W}$</tex>, the smallest area and lowest power reported to date for ECoG recording (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{BW} \leq 1 \text{kHz}$</tex>). Verified with CHB-MIT [1] and ETHZ [2] databases, the digital backend (DBE) occupies <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.075 \text{mm}^{2}$</tex>, consumes <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$6.76 \mu \mathrm{W}$</tex> and achieves event-based sensitivity and specificity of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$94.93 \% / 98.58 \%$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$99.55 \% / 99.97 \%$</tex>, respectively, in patient-specific scenarios.