Abstract
This paper describes a scalable, highly connected, 3-D optoelectronic neural system that uses freespace optical interconnects with silicon-VLSI based hybrid optoelectronic circuits. The system design uses an efficient combination of pulse-width modulating optoelectronic neurons and pulse-amplitude modulating electronic synapses. A prototype system is built and applied to a simple classification problem. An optoelectronic testbench for evaluating learning algorithms suitable for the optoelectronic architecture is implemented. Section 2 briefly describes the hardware requirements for learning neural networks. The fully connected optoelectronic neural architecture is presented in Section 3. The neural system design, including the optical system design, the selection of optimum data encoding methods, and the neuron and synapse circuit designs is presented in section 4. An 8x8 synapse prototype of the optoelectronic neural system is described in section 5. A modification to the architecture that allows an efficient parallel implementation of error backpropagation learning is presented is section 6. The chip-inloop test-bench for learning algorithms is described in section 7. In section 8, future directions for the optoelectronic architecture are discussed; these include limited interconnect neural systems and parallel weight loading that allow receptive fields of arbitrary sizes and connection multiplexing to be achieved.
Original language | English (US) |
---|---|
Pages (from-to) | 416-436 |
Number of pages | 21 |
Journal | Proceedings of SPIE - The International Society for Optical Engineering |
Volume | 2026 |
DOIs | |
State | Published - Nov 9 1993 |
Externally published | Yes |
Event | Photonics for Processors, Neural Networks, and Memories 1993 - San Diego, United States Duration: Jul 11 1993 → Jul 16 1993 |
ASJC Scopus subject areas
- Electronic, Optical and Magnetic Materials
- Condensed Matter Physics
- Computer Science Applications
- Applied Mathematics
- Electrical and Electronic Engineering