Neuromorphic CMOS circuits implementing a novel neural segmentation model based on symmetric STDP learning
福田 駿2006 年度 卒 ／学士（工学）
Humans can distinguish multiple sensory sources that coincide. Recent discoveries of synchronous oscillations in the visual and auditory cortex have triggered much interest in exploring oscillatory correlation to solve the problems of neural segmentation. Many neural models that perform segmentation have been proposed, but they are often difficult to be implemented on practical integrated circuits. Therefore in my study, I designed a simple neural segmentation model that is suitable for analog complementary metal-oxide-semiconductor (CMOS) circuits.
My segmentation model consists of mutually-coupled neural oscillators exhibiting synchronous (or asynchronous) oscillations. All the neurons are coupled with each other through positive or negative synaptic connections. Each neuron accepts external inputs, e.g., sound inputs in the frequency domain, and oscillates (or does not oscillate) when the input amplitude is higher (or lower) than a given threshold value. Our basic idea is to strengthen (or weaken) the synaptic weights between synchronous (or asynchronous) neurons, which may result in phase-domain segmentation. The synaptic weights are updated based on symmetric spike-timing dependent plasticity (STDP) using a correlation neural network that is suitable for analog CMOS implementation. I numerically demonstrated basic operations the proposed model, as well as fundamental circuit operations using a simulation program with integrated circuit emphasis (SPICE).