OPIE

The seduction of neural models for computation is that unlike traditional computers, which are explicitly programmed and meticulously debugged, neuro-computers acquire their behavior by being trained. That is, when stimulated in some manner they are iteratively adjusted until the system exhibits the desired response or behavior. Such machines offer the potential to acquire behavior from experience - to grow smarter with age. OPIE is an optical neuro-computer. It was one of the first real implementations of what has become known as the neural net model of computation. Neural nets, once a controversial approach to artificial intelligence, are now used in an ever-growing number of real-world applications.

Before OPIE, neuro-computations were performed on traditional digital computers by creating a numeric model of the neural network's interconnection matrix. These systems function by propagating a numeric model of input stimulation, updating the computer's internal model of the network, and finally providing some method for presenting the output response of the network, thus achieving simulated neuro-computation.

OPIE was real in the sense that instead of creating a numeric model of synapses, neurons and connectivity, its essential elements were constructed using specially fabricated solid-state devices. OPIE used an array of optically responsive solid-state devices capable of mimicking the behavior of neural net synapses. OPIE was the largest functioning neuro-computer built with 120 neurons fully interconnected via its synaptic array. When stimulated, the machine would electronically settle into a stable computational configuration equivalent to the solutions obtained from artificial simulated neural networks.

Each synapse in this machine is an optically sensitive silicon device similar to a solar cell. The machine's behavior is determined by light falling onto an array of these optical synaptic elements. Since each synapse operates independently, the machine is inherently parallel. By some estimates, this type of computer approaches the theoretical limits of computation (in both speed and power). A very stylish graphical user interface was developed to enable users to train the computer and also to watch the computer cycle toward convergence.

OPIE is a unique computer architecture for many reasons. It is both electronic AND optical. It mixes digital and analog electronics. The computer's "program" consists of optically reconfiguring and adjusting the synapses in its matrix: a process we sometimes termed "enlightenment". Theoretically it is a solar powered computer. It is a very different set of ideas for building interesting computers.

Although originally designed to perform neural style computation, its repertoire of useful computational abilities has grown with time and experience. I will describe a variety of interesting and traditional problems that it can solve.

OPIE was showcased by AT&T at the 1987 Comdex Electronic Trade Show as the first, functioning optical computer capable of solving neural-style computations. Each year one project within the research division of Bell Labs was chosen to represent Bell Laboratories. OPIE was featured in a number of business, news, and technical publications.