The mixed-signal design for a silicon Cort-X: A pathway to an artificial brain
Abstract
The cerebral cortex has been the subject of intense multidisciplinary research, because it is believed to be the seat of all high-level brain functions. Based on previous research in different fields, a modelling architecture is developed for a cortical region. A processing element in the network is developed to model the dynamics of a cortical column. The system level simulation of the network shows interesting dynamics similar to its biological counterpart. As a further effort, the network is implemented into CMOS IC, which is named Cort-X. ^ A mixed-signal framework for the Cort-X is studied. The design of a high-speed, high-resolution and low-power A/D converter is essential for the Cort-X, and for any large-scale parallel processing system. A nonlinear background calibration architecture is developed for the pipeline ADC. The scheme is able to push the performance envelope of pipeline ADCs further outwards, with a calibration process at the background and small calibration overhead. A prototype pipeline ADC is fabricated in a 0.6um CMOS process. After calibration, the ADC is able to sample at 50MS/s, with 12-bit resolution. ^ Two analog Cort-X networks are developed. Cort-X I is designed in a 0.6/μm CMOS process. The Cort-X network is highly nonlinear. The design uses a self-calibration technique to obtain nonlinearity and to compensate for process variations. Cort-X II is a power-aware 0.25μm CMOS design. By extensively using feed-forward open-loop circuits and new architectures, Cort-X II reduces the power consumption by 20 times. New self-calibration techniques are developed to improve the open-loop accuracy. Cort-X II includes 16 processing elements. The chip is designed to be highly scalable to form a network with large processing capability. ^
Subject Area
Engineering, Electronics and Electrical
Recommended Citation
Jie Yuan,
"The mixed-signal design for a silicon Cort-X: A pathway to an artificial brain"
(January 1, 2006).
Dissertations available from ProQuest.
Paper AAI3225569.
http://repository.upenn.edu/dissertations/AAI3225569