We have developed a complementary pair of pFET and nFET floating-gate
silicon MOS transistors for analog learning applications. The memory s
torage is nonvolatile; hot-electron injection and electron tunneling p
ermit bidirectional memory updates. Because these updates depend on bo
th the stored memory value and the transistor terminal voltages, the s
ynapses can implement a learning function. We have derived a memory-up
date rule for both devices, and have shown that the synapse learning f
ollows a simple power law. Unlike conventional EEPROMs, the synapses a
llow simultaneous memory reading and writing. Synapse transistor array
s can therefore compute both the array output, and local memory update
s, in parallel. We have fabricated prototype synaptic arrays; because
the tunneling and injection processes are exponential in the transisto
r terminal voltages, the write and erase isolation between array synap
ses is better than 0.01%. The synapses are small, and typically are op
erated at subthreshold current levels; they will permit the developmen
t of dense, low-power silicon learning systems.