An Artifact-Free 290${\mu}$m²/ch 610nW/ch Neural Readout Frontend With Hybrid EDO Compensation for High-Channel-Count...

Area of Science:

• Biomedical Engineering
• Neurotechnology
• Integrated Circuit Design

Background:

• Next-generation neurorehabilitation requires high-channel-count, closed-loop systems with minimal area and power consumption.
• Existing systems struggle to record neural signals during stimulation due to large, saturating artifacts, or incur excessive overhead for artifact tolerance.

Purpose of the Study:

• To introduce a paradigm shift in readout frontend design for ultra-compact and artifact-tolerant neural signal acquisition.
• To enable high-fidelity neural feature extraction even during stimulation pulses.

Main Methods:

• Developed an ultra-compact, artifact-tolerant readout frontend permitting brief saturation and using backend signal reconstruction via interpolation.
• Implemented a time-multiplexed readout for 64 inputs using a second-order fully time-based incremental analog-to-digital converter.
• Utilized hybrid electrode offset compensation to minimize area overhead.

Main Results:

• Achieved a state-of-the-art 290 μm²/ch area occupation and 610 nW/ch power consumption in a 40nm CMOS process.
• Demonstrated artifact tolerance validated in saline, with feature extraction error below ±10% even under harsh stimulation.
• Successfully extracted high-fidelity neural features with minimal error through signal interpolation.

Conclusions:

• The proposed readout frontend offers a breakthrough in achieving ultra-compact, low-power, and artifact-tolerant neural recording for advanced neurorehabilitation.
• This technology facilitates applications like multi-type epileptic seizure detection, brain-machine interfaces, and brain-to-text conversion.
• The hybrid approach effectively balances area, power, and signal fidelity for next-generation neural implants.