Jian Xu

Adaptive closed-loop neuromodulation is an emerging therapeutic paradigm for treating neurological and psychiatric diseases. However, its hardware implementation remains challenged by limited regulation accuracy, high latency & energy overhead, and poor cross-workload compatibility. To address these faced challenges, this article presents a multi-task neural processing system-on-chip (SoC) with three key technologies. First, an established multi-input multi-output linear state-space model (MIMO LSSM) with linear quadratic Gaussian (LQG) control is configured for deterministic on-chip execution to improve regulation accuracy. Second, a processing element (PE)-array-aware compact parameter-encoding scheme is proposed to reduce storage cost and memory-access overhead. Third, a mode-configurable compute fabric (MCCF) is designed to support diverse neuromodulation workloads on a unified hardware fabric. The designed SoC was fabricated in a TSMC 65nm CMOS process. Measured results and performance comparison show that it achieves a maximum energy efficiency of 1.43 TOPS/W (3.08×), a maximum area efficiency of 1.09 GOPS/mm2 (13.29×), and a peak performance of 5.12 GOPS (40.96×). Besides, the SoC has been demonstrated on the BONN and DEAP datasets, achieving accuracies of 99.18% in seizure detection and 92.3% in emotion detection. Overall, the proposed SoC offers a competitive hardware solution for adaptive closed-loop neuromodulation.

High-precision implantable multi-channel neural recording systems are considered as having a crucial role in the diagnosis and treatment of neurological disorders. However, it is a significant design challenge to achieve an optimal trade-off among linear parameters, signal fidelity, power consumption, and circuit area. To address this challenge, a Simulink-based modeling approach has been proposed to incorporate adjustable non-linear parameters across the front-end circuits and analog-to-digital converter (ADC) stages. The model evaluates non-linearity impacts on system performance through both quantitative spike detection accuracy analysis and a neural decoding paradigm based on Chinese handwriting reconstruction. Simulated results show that total harmonic distortion (THD) can be set to −34.32 dB for the low-noise amplifier (LNA), −33.73 dB for the programmable gain amplifier (PGA), and −57.95 dB for the ADC in order to achieve reliable detection accuracy with minimal design cost. Moreover, ADC non-linearity has a greater influence on system performance than that of the LNA and PGA. The proposed approach offers quantitative and systematic hardware design guidance to balance signal fidelity and resource efficiency for future low-power, high-accuracy neural recording systems.

This paper presents a highly-integrated single-link wireless neuromodulation System-on-Chip (SoC) for compact injectable implanted medical devices (IMDs). The SoC integrates a single-link wireless power data transfer (WPDT) module, eliminating the need for bulky batteries and wire connection. A dual-freshened 4-phase switching scheme is developed to realize a parasitic-insensitive capacitive-DAC-based stimulator, which ensures a 5 μA -10 mA current range for multiple therapeutic applications. Besides, a switch-capacitor ladder charge-pump is adopted to provide 15 V voltage compliance. An impedance estimation module with high-side voltage sensing capability is integrated for checking electrode contact. Fabricated in a 180-nm Bipolar-CMOS-DMOS(BCD) high voltage (HV) process, the SoC occupies an area of 2.1 mm × 4.8 mm. Measurement results and in vivo experiment verify the feasibility and efficacy of the proposed SoC and demonstrate that the proposed SoC is suitable for minimally invasive injectable implants.

Closed-loop deep brain stimulation (DBS) is an emerging technology in the field of neuromodulation, which can provide effective treatment methods for intractable neurological disorders. To address the faced challenges, this paper presents a high-precision closed-loop DBS device based on a custom-designed neuromodulation integrated circuit fabricated in a 180 nm Bipolar-CMOS-DMOS (BCD) process. Also, three new techniques have been developed to optimize system performance. Firstly, to improve system stability, a frequency shaping & quick blanking amplifier (FSQBA) based analog front-end is proposed to extend system dynamic range (DR). Secondly, both a stimulation artifacts (SA) suppression scheme and a closed-loop control algorithm are developed to achieve real simultaneous recording and stimulation. Thirdly, a redundant-crossfire (RXF) technique is designed to enhance current precision without consuming much extra area. Bench-top measurements show that the designed system features a wide DR of 112 dB, a short saturation recovery time of 100 μs, and a high stimulation current resolution of 13.1-bit. Besides, animal experiments demonstrate that the designed system is highly suitable for long-term implantation applications, thus providing accurate and reliable closed-loop control for neuromodulation therapies.

Electrical stimulation is an important technique for modulating the functions of the nervous system through electrical stimulus. To implement a more competitive prototype that can tackle the domain-specific difficulties of existing electrical stimulators, three key techniques are proposed in this work. Firstly, a load-adaptive power saving technique called over-voltage detection is implemented to automatically adjust the supply voltage. Secondly, redundant digital calibration (RDC) is proposed to improve current accuracy and ensure safety during long-term electrical stimulation without costing too much circuit area and power. Thirdly, a flexible waveform generator is designed to provide arbitrary stimulus waveforms for particular applications. Measurement results show the stimulator can adjust the supply voltage from 12 V to 100 V automatically, and the measured effective resolution of the stimulation current reaches 14 bits in a full range of 6.5 mA. Without applying charge balancing techniques, the average mismatch between the cathodic and anodic current pulses in biphasic stimulus is 0.0427%. The proposed electrical stimulator can generate arbitrary stimulus waveforms, including sine, triangle, rectangle, etc., and it is supposed to be competitive for implantable and wearable devices.

This paper presents a high precision, wide dynamic range (DR) closed-loop neuromodulation (CLNM) system that can completely reject stimulation artifacts (SA) and achieve rapid SA recovery. In the recorder, a novel SA quick-blanking scheme is proposed for rail-to-rail SA rejection while minimizing SA recovery time. Besides, a new analog front-end (AFE) architecture based on a frequency-shaping (FS) technique is developed to extend DR intrinsically. In the stimulator, a stimulation driver implemented with a proposed redundant crossfire (RXF) technique is incorporated to improve the effective resolution of the stimulation current. The designed CLNM system is implemented in a 180nm Bipolar-CMOS-DMOS (BCD) process. Measurement results show that the system is capable of tolerating rail-to-rail (5V) SA and reducing the SA recovery time from 12 ms to 0.15 ms. The FS recorder extends the DR at low frequencies (LF) to 17.5 bits to enhance tolerance to LF interferences. The proposed stimulator adopting the 4-way RXF topology improves the effective resolution to 12.75 bits without consuming much extra area and power. Animal experiments demonstrate that the designed system can acquire high-fidelity neural signals immediately after stimulation onsets, thus supporting concurrent recording and stimulation.

Concurrent recording of neural activities and functional magnetic resonance imaging (fMRI) data is useful for studying the neurovascular coupling relationship. This article presents a low-noise, frequency-shaping based neural recorder chip that is insensitive to radio frequency (RF) pulses and gradient echo artifacts under strong magnetic environment. To support simultaneous recording of local field potentials (LFPs), extracellular spikes, and fMRI data, a magnetic resonance imaging (MRI) compatible data acquisition (DAQ) device based on the designed recorder chip is developed with multiple circuit optimization techniques. Bench-top measurement shows that the designed DAQ device has 4.5 $\mu$ V input-referred noise integrated from 300 Hz to 3000 Hz, which is not greatly affected by electromagnetic interference (EMI) at ultrahigh magnetic field (UMF, 16.4 T). In animal experiments, the designed DAQ device has been demonstrated to be capable of acquiring both the LFPs and extracellular spikes from a rat's brain before, during, and after MRI scanning. Besides, no obvious artifacts are seen from the designed DAQ device at multiple typical MRI scanning modes, and the system recovery time after gradient artifacts is reduced from more than 25 ms to less than 5 ms. The proposed DAQ device architecture based on the frequency-shaping neural recorder chip is MRI compatible and can provide highly competitive performance for concurrent recording of neural activities and fMRI data.

Objective: The next generation prosthetic hand that moves and feels like a real hand requires a robust neural interconnection between the human minds and machines. Methods: Here we present a neuroprosthetic system to demonstrate that principle by employing an artificial intelligence (AI) agent to translate the amputee’s movement intent through a peripheral nerve interface. The AI agent is designed based on the recurrent neural network (RNN) and could simultaneously decode six degree-of-freedom (DOF) from multichannel nerve data in real-time. The decoder’s performance is characterized in motor decoding experiments with three human amputees. Results: First, we show the AI agent enables amputees to intuitively control a prosthetic hand with individual finger and wrist movements up to 97-98% accuracy. Second, we demonstrate the AI agent’s real-time performance by measuring the reaction time and information throughput in a hand gesture matching task. Third, we investigate the AI agent’s long-term uses and show the decoder’s robust predictive performance over a 16-month implant duration. Conclusion & significance: Our study demonstrates the potential of AI-enabled nerve technology, underling the next generation of dexterous and intuitive prosthetic hands.

Objective. Deep learning-based neural decoders have emerged as the prominent approach to enable dexterous and intuitive control of neuroprosthetic hands. Yet few studies have materialized the use of deep learning in clinical settings due to its high computational requirements. Approach. Recent advancements of edge computing devices bring the potential to alleviate this problem. Here we present the implementation of a neuroprosthetic hand with embedded deep learning-based control. The neural decoder is designed based on the recurrent neural network architecture and deployed on the NVIDIA Jetson Nano—a compacted yet powerful edge computing platform for deep learning inference. This enables the implementation of the neuroprosthetic hand as a portable and self-contained unit with real-time control of individual finger movements. Main results. A pilot study with a transradial amputee is conducted to evaluate the proposed system using peripheral nerve signals acquired from implanted intrafascicular microelectrodes. The preliminary experiment results show the system’s capabilities of providing robust, high-accuracy (95%–99%) and low-latency (50–120 ms) control of individual finger movements in various laboratory and real-world environments. Conclusion. This work is a technological demonstration of modern edge computing platforms to enable the effective use of deep learning-based neural decoders for neuroprosthesis control as an autonomous system. Significance. The proposed system helps pioneer the deployment of deep neural networks in clinical applications underlying a new class of wearable biomedical devices with embedded artificial intelligence. Clinical trial registration: DExterous Hand Control Through Fascicular Targeting (DEFT). Identifier: NCT02994160.