No abstract

losed-loop neuromodulation improves open-loop therapeutic electrical stimulation by providing adaptive, on-demand therapy, reducing side effects and extending battery life in wireless devices 1,2. Closing the loop requires low-latency extraction and accurate estimation of neural biomarkers 3-5 from recorded signals to automatically adjust when and how to administer stimulation as feedback to the brain. Recent studies have shown responsive stimulation to be a viable option for treating epilepsy 2,6 , and there is evidence that closed-loop strategies could improve deep brain stimulation (DBS) for treating Parkinson's disease and other motor disorders 7,8. However, there is presently no commercial device allowing closed-loop stimulation for DBS in patients with movement disorders, and strategies for implementing such stimulation are still under investigation. In fact, most attempts to close the loop for DBS treatments have been done only for short duration using systems that were not fully implantable 4,5,9-11. To enable advanced research in closed-loop neuromodulation, there is a need for a flexible research platform, for testing and implementing these various closed-loop paradigms, that is also wireless, compact, robust and safe. Designing such a device requires unification of multi-channel recording, biomarker detection and microstimulation technologies into a single unit with careful consideration of their interactions. Wireless, multi-channel recording-only devices capture activity from wide neuronal populations 12,13 , but do not have the built-in ability to immediately act on that information and deliver stimulation. Several complete closed-loop devices have been proposed and demonstrated, but are limited by low channel counts 14-17 and low wireless streaming bandwidth 14-18. Most recently, variations of the fully integrated and optimized closed-loop neuromodulation system-on-a-chip (SoC) have been presented, but full system functionality has not yet been adequately demonstrated in vivo 19-25. While future versions may be paired with miniaturized external battery packs and controllers, current systems built around these SoCs require large, stationary devices to deliver power inductively from a close range 19,21,23. This limits studies to using small, caged animals. Furthermore, any device for concurrent sensing and stimulation must be able to mitigate or remove stimulation artefactsthe large voltage transients resulting from stimulation that distort recorded signals and obscure neural biomarkers. Signals recorded concurrently with stimulation may contain relevant information for closed-loop algorithms or offline analysis, yet existing devices disregard these affected windows of data, or fail to reduce artefacts to an acceptable level for recovery of many potentially useful biomarker features. Effectively and efficiently cancelling artefacts requires careful co-design of the stimulators and signal acquisition chains. Additionally, computational reprogrammability is needed for application-dependent algorithm de...

Closed-loop and responsive neuromodulation systems improve open-loop neurostimulation by responding directly to measured neural activity and providing adaptive, on-demand therapy. To be effective, these systems must be able to simultaneously record and stimulate neural activity, a task made difficult by persistent stimulation artifacts that distort and obscure underlying biomarkers. To enable simultaneous stimulation and recording, several techniques have been proposed. These techniques involve artifact-preventing system configurations, resilient recording front-ends, and back-end signal processing for removing recorded artifacts. Co-designing and integrating these artifact cancellation techniques will be key to enabling neuromodulation systems to stimulate and record at the same time. Here, we review the state-of-the-art for these techniques and their role in achieving artifact-free neuromodulation.

EMG-based gesture recognition shows promise for human-machine interaction. Systems are often afflicted by signal and electrode variability which degrades performance over time.We present an end-to-end system combating this variability using a large-area, high-density sensor array and a robust classification algorithm. EMG electrodes are fabricated on a flexible substrate and interfaced to a custom wireless device for 64-channel signal acquisition and streaming. We use braininspired high-dimensional (HD) computing for processing EMG features in one-shot learning. The HD algorithm is tolerant to noise and electrode misplacement and can quickly learn from few gestures without gradient descent or back-propagation. We achieve an average classification accuracy of 96.64% for five gestures, with only 7% degradation when training and testing across different days. Our system maintains this accuracy when trained with only three trials of gestures; it also demonstrates comparable accuracy with the state-of-the-art when trained with one trial.

We present an 180nm HV CMOS IC for concurrent neural stimulation and recording that combines 64 low-noise recording front-ends and 4 independent stimulators multiplexed to any of the 64 channels. The stimulators have 5mA peak current, 12V compliance and dynamic power management to maximize efficiency. Co-design of the stimulation and recording subsystems resulted in 100mV of recording linear range, 70nV/rtHz noise, and a rapid 1ms (single-sample) artifact recovery during stimulation.

No abstract

No abstract

Varying contraction levels of muscles is a big challenge in electromyography-based gesture recognition. Some use cases require the classifier to be robust against varying force changes, while others demand to distinguish between different effort levels of performing the same gesture. We use braininspired hyperdimensional computing paradigm to build classification models that are both robust to these variations and able to recognize multiple contraction levels. Experimental results on 5 subjects performing 9 gestures with 3 effort levels show up to 39.17% accuracy drop when training and testing across different effort levels, with up to 30.35% recovery after applying our algorithm.