Self-learning chips to implement many popular ANN (artificial neural networ
k) algorithms are very difficult to design. We explain why this is so and s
ay what lessons previous work teaches us in the design of self-learning sys
tems. We offer a contribution to the "biologically-inspired" approach, expl
aining what we mean by this term and providing an example of a robust, self
-learning design that can solve simple classical-conditioning tasks, We giv
e details of the design of individual circuits to perform component functio
ns, which can then be combined into a network to solve the task. We argue t
hat useful conclusions as to the future of on-chip learning can be drawn fr
om this work.