Advanced interconnect technologies in the era of cognitive computing

Abstract

Cognitive computing is capable of machine learning, recognition and proposal. It has a great potential to make human life richer, more productive and more intelligent. For the realization of the cognitive computing, an efficient and scalable non-von Neumann architecture inspired by the human brain structure has been developed and a device which demonstrates the concept was also built. This device mimics the signal processing of the human brain, packing one million neuron circuits in 4,096 cores. It consumes almost 1,000 times less energy per event compared with a state-of-the-art multiprocessor. However, one million neurons only correspond to those of the bee's brain, and the inter-chip wiring becomes a key element to mimic the brains of higher order animals because this kind of neuromorphic device requires a large number of parallel signal lines for massive parallel signal operations. 3D chip stacking is one of the prospective technologies in achieving the device. In 3D chip stacking, technologies such as low cost TSV formation and fine-pitch interconnection, smaller than 10μm pitch technology are required. From the device reliability point of view, the optimization of solder composition for fine-pitch interconnection is also important. Injection Molded Solder (IMS) is well fit to this fine pitch interconnection, in terms of material optimization and low cost joints. As for the interposer, the build-up organic interposer is the most attractive candidates for the cost issue, but in the most top layer, ultra-fine pitch wiring with the line and space widths smaller than 1μm should be prepared. A number of material and process innovations are necessary for the inter-chip connection of neuromorphic devices.