Wafer-Scale Integration (WSI)
Wafer-Scale Integration (WSI) is a rare and advanced method of semiconductor integration that allows for the fabrication of very large, multi-chip circuits on a single wafer. This technology enables faster and more efficient processing than conventional chip packaging methods by allowing multiple chips to be interconnected with high-speed connections on a single substrate, as well as improving signal integrity and reducing power consumption. WSI also eliminates the need to package individual components, saving space and cost.
Wafer-scale integration is a way to bypass the main obstacle in achieving higher computer performance: off-chip communication. Sending data between chips requires electrical signals along metal wires, which are slow and energy inefficient due to their length and thickness. The number of connection points at chip boundaries can also be limiting when it comes to moving data quickly. Wafer-scale integration offers more room for compute cores, fast local memory, and efficient on-chip communication compared with separate packaged chips; providing over 50 times more area than conventional chips.
Many companies in the 1970s and 1980s attempted to develop WSI production systems but failed. Texas Instruments, ITT Corporation, Trilogy Systems, Digital Equipment Corporation, and Anamartic were some of the main competitors who attempted to develop this technology. However the effort was plagued by a multitude of issues. The last attempt was in 1989 when Anamartic was unable to ensure a large enough supply of silicon wafers and succumbed to financial difficulties in 1992.
On August 19, 2019, Cerebras Systems unveiled their development progress of WSI for deep learning acceleration in the form of Cerebras Wafer-Scale Engine (WSE) which allows for full integration of compute, memory, and I/O on a single wafer. Cerebras’ industry-leading innovation has enabled the development of powerful new AI products and continues to revolutionize the way we think about computing.
