The goal of FOCUS was to define device and system requirements for neural-type networks. The objectives were to:
-Develop analytical and numerical models of neural-type networks that will allow optoelectronic devices to be specified for their implementation.
-Study the properties of the various state-of-the-art optoelectronic devices, particularly those based on III-V multiple quantum well (MQW) materials, oriented towards their use in neural network applications.
-Study the technologies for integrating such processing devices with silicon electronic processing elements.
-Exchange information with other programmes and projects on these and related devices and system studies.
The opportunities for optoelectronic technologies in information technology were examined, especially in the field of neural type networks. Expertize in the design, growth and fabrication of III to V optoelectronic devices, the design of advanced information processing systems, computer architectures, and general digital and analogue systems, design skills were brought to bear on this task.
An in depth understanding of the design and optimization of PIN-multiple quantum well (MQW) electroabsorption modulator structures has been built up drawing upon data from the partners for a wide range of II V materials systems. New nonlinear optical properties have also been observed and collaborative studies of new techniques for integrating such devices with electronic circuits for use in interconnects have been carried out.
Novel optical implementations of neural type networks have been demonstrated and advances have been made in the understanding of the relationship between neural and conventional associative memories that point to new ways in which 'optical learning' might be implemented. These have been tested by numerical simulation.
APPROACH AND METHODS
The work has been broadly split between systems and devices with sub-working groups established to coordinate activities.
In the field of devices the subgroups are on:
-study of strain layer III-V modulators in InGaAs
-design of modulators and detectors in III-V MQW structures
-integration of III-V devices with silicon circuitry
-waveguiding devices in III-V MQW materials
-non-linear optical properties of III-V MQW structures
-characterisation of MQW materials.
In the field of systems:
-theoretical modelling of optical implementations of neural-type networks
-study of the application of optical neural-type networks to switching systems
-system sensitivity to device characteristics.
PROGRESS AND RESULTS
An in-depth understanding of the design and optimisation of PIN-MQW electro-absorption modulator structures has been built up drawing upon data from the partners for a wide range of II-V materials systems. New non-linear optical properties have also been observed and collaborative studies of new techniques for integrating such devices with electronic circuits for use in interconnects have been carried out.
Novel optical implementations of neural-type networks have been demonstrated and advances have been made in the understanding of the relationship between neural and conventional associative memories that point to new ways in which "optical learning" mightbe implemented. These have been tested by numerical simulation.
It has long been clear that optics brings a powerful interconnect capability to electronics and PIN-MQW devices are prime candidates for use as opto-electronic interface elements. In the short term, this interconnect capability is most likely to be benefi cial in multiprocessor machines and in very high data throughput processors such as telecommunication switches. Neural networks rely upon very rich and complex interconnection patterns for their effectiveness, and optics has a major role to play in their implementation. This Action has studied some of the device requirements for such networks and a range of relevant device technologies.
SO51 0ZN Romsey
WC1E 7HX London