| IP개요 |
This work presents a hardware Liquid State Machine (LSM) optimized for speech classification using a high-linearity body-driven VCO-based LIF neuron. The proposed neuron operates under 0.6V, leveraging phase-domain integration. Integration linearity is further enhanced through body-driven VCO and a differential structure. A resistor-based synapse is proposed, featuring 4-bit programmable weights and dummy inputs to equalize time constants across neurons. The LSM consists of 36 reservoir neurons and 238 synapses, targeting 88.0% classification accuracy for spoken digit recognition. The recurrent neurons will be fabricated in Samsun 28nm CMOS process. |
