| Тойм: | When the need for cost saving cognitive computing keeps increasing, the conventional von
Neumann architecture approaches its limitations. Neuromorphic computing architecture is
inspired by the brain and has been widely explored during the past three decades due to its
energy efficiency and massive connectivity. A critical component of the brain-inspired
architecture is the synapse, which transmit electrical signals between neurons. Many
devices such as memristors and phase-change memory have been researched as possible
candidates for analog synapses. In this report, a CMOS-only and manufacturing-ready
device- the transistor, is researched.
This paper presents a study of the transistor on how the analog characteristics of the
transistor can be used for neuromorphic applications by showing how the transistor can be
used to simulate an analog synapse. Next, an algorithm for unsupervised learning, the
Winner-Takes-All (WTA) clustering, are researched using the transistors as analog
synapses. A software simulation of the neuromorphic computer built using transistors as
analog synapses will test on the feasibility of hardware implementation. System
performance will be evaluated using experimentally obtain transistor characteristics.
|
|---|