Technology from the 80s: who revives waferscale processors

    The idea of ​​waferscale processors made a lot of noise in the 80s of the last century, but turned out to be forgotten for decades due to the complexity of production. Today, technology is again trying to develop. We tell who works in this area. / photo timo vn CC BY

    What is waferscale

    In the production of modern processors, the silicon substrate is cut into individual crystals, from which the chip is then assembled. Waferscale (WSI) technology implies the creation of processors (both CPU and GPU) on a single semiconductor wafer - it becomes one big "chip". This approach increases system performance: components are placed closer, so data is transferred faster.

    In the 1970s and 1980s, WSI technology was proposed to be used to create chips for supercomputers. However, it was decided to abandon this idea - there were too many defects in the chips during production.

    For example, the Trilogy Systems startup, which then became the most funded in Silicon Valley, worked in this area - it received $ 230 million in investments. But the company was not able to form a reliable process technology. In 1985, Trilogy discontinued WSI research and went bankrupt, which negatively impacted the overall reputation of waferscale processors.

    Only recently, WSI was again talked about as a promising area. For example, it is expected that such devices will help increase the productivity of IT infrastructure in data centers. Interfaces for transferring data between server components often turn out to be a bottleneck. Waferscale-systems, by placing part of the components on one substrate, can help solve the problem of bandwidth.

    Who develops technology

    Organizations developing watercale processors today are the University of California at Los Angeles and the University of Illinois at Urbana-Champaign. In early February of this year, university staff proposed creating a system that combines dozens of GPUs on a single silicon. The project is called Silicon Interconnect Fabric, or Si-IF.

    Engineers managed to create two prototypes for 24 and 40 GPUs. The developers conducted comparative tests with classical devices (with the same number of computing modules) - the performance of Si-IF devices was 2.5–5 times higher. Potentially, the technology can accelerate the work of servers based on GPUs in data centers.

    Since 2012, Taiwanese manufacturer TSMC has also been involved in Waferscale processors. Their technology is called CoWoS - "chip on a wafer on a substrate." On the silicon bridge (interposer) are the GPU and memory, FPGA and memory or network controllers. Then the bridge is mounted on the circuit board and "packaged" like a regular processor. The solution is already used in real devices, for example, the Nvidia Volta series graphics chips.

    The technology is also being developed by engineers as part of the WIPE project, which is part of the European Horizon 2020 innovative solutions initiative . WIPE has seven members, including IBM, the Belgian Center for Microelectronics iMinds and a number of research institutes.

    / a photoEnrique Jiménez CC BY-SA

    Organization proposes to create a processor for working with fiber optic networks. In it, a module for converting an optical signal into an electric one will be located above the silicon wafer with chips and connected to it directly through metal contacts. Researchers expect such an architecture to consume less power and run faster than traditional systems with separate transceivers and processors.

    What are the prospects

    Experts expect that the development of WSI processors will help data centers increase server capacity for demanding tasks, such as training neural networks. According to some reports, the Cerebras startup is already working on a waferscale processor for machine learning - although the company itself hides the development details.

    Another advantage of WSI is its ease of management. For operating systems and other programs, waferscale devices do not look like several separate processors, but, for example, like one large GPU. This simplifies the work of programmers and system administrators who do not need to individually configure each module of the waferscale system.

    But waferscale processors have a limitationwhich makes it difficult to scale IT infrastructure. Power systems in WSI chips require the installation of voltage regulators. These blocks occupy a place on which it would be possible to place additional computing modules. Another limitation for waferscale processors is related to device cooling. The high density of the modules on the chip complicates heat dissipation.

    But according to experts, these problems are surmountable and now is the right time to revive WSI. Since the 1980s, processor manufacturing technologies have become more advanced, the number of defects has decreased. Therefore, we can expect that in the near future more and more projects will appear on the development of waferscale devices.

    What we write about in a corporate blog:

    And in the Telegram channel:

    Also popular now: