A frequency-domain scheme to share a communications channel among the cells of a CNN is proposed. The scheme is based on a modification of the wave-parallel-computing technique and addresses the problem of reducing the number of communication links. Reduction in the communication paths is achieved by frequency multiplexing. This makes it possible to have simultaneous full-parallel access to all the cells of the array. The approach also takes advantage of the parallelism inherent in wave-parallel-computing to solve part of the state equation within the same channel during a transmission operation. Moreover, with this architecture, the CNN array is not required to have a physical matrix array of cells, providing in this form even more flexibility for the hardware implementation. A system level simulation was done and operating ranges were found as an aid to propose a final system architecture of a tentative VLSI IC.
[1]
Lin-Bao Yang,et al.
Cellular neural networks: theory
,
1988
.
[2]
Takafumi Aoki,et al.
Wave-parallel computing technique for neural networks based on amplitude-modulated waves
,
1996,
Proceedings of 26th IEEE International Symposium on Multiple-Valued Logic (ISMVL'96).
[3]
A. Murray.
Pulse techniques in neural VLSI: a review
,
1992,
[Proceedings] 1992 IEEE International Symposium on Circuits and Systems.
[4]
Michael P. Craven,et al.
Consideration of multiplexing in neural network hardware
,
1994
.
[5]
José Pineda de Gyvez,et al.
Time-Multiplexing Scheme for Cellular Neural Networks Based Image Processing
,
1996,
Real Time Imaging.
[6]
L.O. Chua,et al.
Cellular neural networks
,
1993,
1988., IEEE International Symposium on Circuits and Systems.