Modern artificial intelligence increasingly operates at the edge, where sensing, computation, memory, and learning must function under strict constraints on power, latency, and adaptability. This challenge is especially important for biological signals such as EMG and EEG, which are weak, noisy, time-varying, and highly user-dependent. Conventional systems rely on continuous sampling, dense processing, and repeated data movement between separate memory and processor units, leading to unnecessary activity and the von Neumann bottleneck. This talk introduces neuromorphic computing as an alternative based on event-driven processing, spiking dynamics, in-memory computation, and adaptive physical synapses. Using two case studies from my work, I will show how EMG signals can be processed through event-based spiking reservoirs, and how EEG decoding can be implemented with ferroelectric memristive synapses using device-aware on-device learning and personalization. Together, these examples illustrate how efficient biosignal intelligence requires co-design across events, dynamics, memory, learning, and material physics.