(15) Design and testing of a multiplexed neural interface

Introduction: Neural interfaces are a critical part of BCIs and serve to bridge the gap between the brain and electronics. A tradeoff ubiquitous across neural interfaces is that of spatial coverage and resolution as the number of recording channels is limited by connector sizes and backend electronics. An approach to overcome this limitation is to embed electronics at the electrode site, which enables multiple electrodes to share one output wire through time division multiplexing (TDM), thus increasing the number of channels without increasing the number of wires. However, a drawback of TDM is increased noise due to aliasing, and fabrication-dependent characteristics, which contribute to transistor noise.

Methods: In this work we combined integrated circuit (IC) design with transfer printing to create an intracortical device with front-end active electronics at the electrode site to enable TDM with low noise. We tested the device pre- and post- transfer printing in vitro and in vivo using auditory stimuli to probe auditory evoked responses in the rat primary auditory cortex (A1). We identified a location of A1 responsive to auditory stimuli using a 61-channel μECoG array and implanted the intracortical device in this location at various cortical depths.

Results: In vitro results demonstrated circuit performance matched simulations, and that transfer printing did not negatively affect performance. The device has a thickness of 37.4 μm (25.4 μm of polyimide and 12 μm of SU-8 encapsulating 8 μm of silicon IC) and 32 electrodes with 50 μm pitch. The front-end pixel filters signals 0.1 Hz - 10.4 kHz , amplifies with 26 dB gain, and multiplexes 32:1 at 10 ksps/ch with input referred noise of 9.18 +/- 3.06 μV (10 - 200 Hz), and 6.95 +/- 1.23 μV (300 - 3000 Hz). Importantly, the pixel architecture occupies an area of 42 μm x 50 μm while consuming < 1 μW of power, enabling 50 μm pitch with a power density below the tissue heating standard of 40 mW/cm^2. In vivo results demonstrated the device could record evoked responses with ESNRs up to 13.7 dB.

Conclusion: Overall, this work demonstrates 1) large regions of ICs can be transfer printed without compromising function to create thin, implantable devices with commercial foundry level circuit performance, and 2) our architecture enables scalable high density, low-noise, and low-power actively multiplexed neural recording. This work can be applied to active implantable devices, and across neural interfaces to create high channel-count arrays and to ultimately improve BCIs.