A neural network approach to high performance analog circuit design

In order to gain new insight into the design of high-precision, high-speed analog circuits, several possible network implementations of an A/D convertor are presented. These networks are marked by programmability and parallelism, which can be used to maintain circuit precision without the use of feedback. This removes design constraints on closed-loop stability, and may lead to faster circuit performance. On-chip training or calibration is likely to be necessary, but can be done in an offline mode, and thus may not hinder circuit speed significantly