An analog standard cell layout configuration is proposed for simplifying the design and reducing the man-hours for designing mixed analog-digital LSIs, and analog standard cells are fabricated for A-D and D-A converters with Δ-Σ modulators. This works seeks to implement 2-D cell placement with up-down and left-right mirror rotation and shorter high-impedance analog wiring than conventional 1-D placement in order to obtain high-performance analog characteristics. By considering sensitivity to noise, routing channels have been classified into 4 types: high-impedance analog, low-impedance analog, analog-digital, and digital, and efforts have been made to prevent analog wires from crossing over digital wires. In addition to power and analog ground wires, analog standard cells have built-in analog ground wires with attached wells optimized for shielding. These wires are interconnected to a new isolation cell that separates analog circuits from digital circuits and routing channels. Based on the above layout structure, 46 different types of analog standard cells have been designed. Also, the analog part of Δ-Σ type A-D and D-A converters can be automatically designed in conjunction with interactive processing and chips fabricated by using these cells. It was found that, compared to manual design, one could easily obtain a chip occupying less than 1.5-times the area with about 2/3 the man-days using this approach. In comparison with manual design, it was also found that the S/N ratio could be reduced from about 6 to 7 dB.

A sparse memory access architecture which is proposed to achieve a high-computational-speed neural-network LSI is described in detail. This architecture uses two key techniques, compressible synapse-weight neuron calculation and differential neuron operation, to reduce the number of accesses to synapse weight memories and the number of neuron calculations without incurring an accuracy penalty. The test chip based on this architecture has 96 parallel data-driven processing units and enough memory for 12,288 synapse weights. In a pattern recognition example, the number of memory accesses and neuron calculations was reduced to 0.87% that needed in the conventional method and the practical performance was 18 GCPS. The sparse memory access architecture is also effective when the synapse weights are stored in off-chip memory.