The increasing number of recording channels in brain-computer interfaces (BCIs) has led to a surge in raw data, making reliable neural spike detection in noisy environments essential for constraining transmission bandwidth. This study introduces a precise and computationally efficient spike detection module that combines stationary wavelet transform (SWT), Teager energy operator (TEO), and root-mean-square (RMS) calculator for adaptive thresholding. The SWT effectively eliminates high-frequency noise, enhancing spike amplification by energy operators, while the hardware implementation is optimized using a lifting scheme and channel-interleaving architecture. Evaluated on the WaveClus datasets, the proposed detector achieves an average accuracy of 98.84%. Implemented in a 65-nm ASIC, the 8-channel spike detector consumes 0.532 μW/Ch and occupies 0.00645 mm²/Ch at a 1.2-V supply voltage. This design offers one of the highest accuracies among state-of-the-art methods while addressing scalability and hardware cost considerations, providing a low-power, area-efficient solution for adaptive spike detection in BCIs.
脑机接口（BCI）中记录通道数量的增加导致了原始数据的激增，这使得在噪声环境中可靠地检测神经尖峰信号成为限制传输带宽的关键。本研究提出了一种精确且计算高效的尖峰检测模块，结合了平稳小波变换（SWT）、Teager能量算子（TEO）和均方根（RMS）计算器，用于自适应阈值处理。SWT有效消除了高频噪声，增强了能量算子对尖峰信号的放大效果，同时通过提升方案和通道交错架构优化了硬件实现。在WaveClus数据集上的评估表明，所提出的检测器平均准确率达到98.84%。在65纳米工艺的专用集成电路（ASIC）中实现时，8通道尖峰检测器在1.2V电源电压下的功耗为0.532 μW/通道，面积为0.00645 mm²/通道。该设计在尖峰检测方法中提供了最高的准确性之一，同时解决了可扩展性和硬件成本的考虑，为BCI中的自适应尖峰检测提供了一种低功耗、面积高效的解决方案。
