Silicon-Interconnect Fabric for Research & Development
Benefits of Silicon-Interconnect FabricSilicon-Interconnect Fabric
The silicone fabric allows the chips to be packed tighter and the heat to be better transported away from the chip and into the fabric rather than out of it.
Get Your Quote FAST!
Silicon-Interconnect Fabric Packageless Processors to Increase PC performance
Researchers from the Department of Electrical Engineering have developed a system made of silicon composites that drives the integration of packageless processors. In a report published Tuesday in the IEEE Spectrum, the researchers suggest that circuit boards can be replaced by silicon, which will help with the size of the building - meaning limited devices. The engineers aim to use a so-called "silicon interconnect fabric" to build a powerful, low-cost and lightweight processor with a wide range of processing capabilities. [Sources: 2, 3, 11]
The silicon connecting tissue dissolves the system - on - a - chip (SoC) into an integrated collection of photonic chips, even if they are physically far apart and can be supplied with electricity via a power distribution network. We could even see photonics chips integrated into multichip modules and connected to photons or analogues in silicon interconnect tissues. [Sources: 2, 6, 9, 11]
One way to do this would be to build a GPU on the same silicon wafer and add a compound, but this is something special: we have a standard GPU chip that has passed quality tests and we have reduced the size of the silicon board from 1,000 cm2 to 400 cm 2. We developed a technology called silicon-interconnect fabric (SiIF) and developed an ultra-thin, high-performance, low-cost printed circuit board (PCB) so thin that it was dubbed the world's first silicon-to-silicon connective fabric. It connects better and is better networked than conventional silicon circuits, such as the standard silicon circuit board. [Sources: 0, 3, 15, 17]
In the silicon space, Cerebras has huge chips that are designed to handle tensor-ops and efficiently transmit data. It has a high-performance, low-cost silicon-to-silicon (SiIF) connective tissue and is super thin. [Sources: 5, 12]
With a technology the team calls Flextrate, the Si-IF is completely flexible, allowing large networks to be assembled in a confined space. Compared to printed circuit boards, it has other advantages: It weighs less and offers an integration approach. As electronics continues to advance in the field of multichip modules, we are looking for tools to work with this new technology in printed circuit boards. We will define an open interface architecture that allows the integration of silicon-to-silicon (SiIF) modules into the silicon nozzle itself. [Sources: 0, 1, 6, 13]
Essentially, the Si-IF replaces the PCB with a silicon substrate and allows the die to be connected directly to the wafer. Think of it as making silicon semiconductors by building everything on a single wafer. [Sources: 4, 6]
This visibility ensures that the right CPU core / FPGA mix is achieved for each Soc, and the visibility of the silicon composite fabric provides high visibility to achieve the "right" CPU core F-PGA mix in the Soc. The Si-IF also allows the packaging to be omitted from the die, since the dimensions, width and pitch of each connecting tissue are similar to those of a chip connector. This is what the FGPA fabric needs to optimize the silicon area, costs and power consumption. [Sources: 8, 10]
This is exactly what the researchers are doing with this new technology, which they call silicon interconnect tissue (SiIF). Instead, it would be like using small manufacturing chips that are tightly bound together in a fabric. Thousands of engineers around the world have done this in the past with silicon microcontrollers, microprocessors, and even silicon transistors. The researchers believe that silicon composite fabrics should allow system designers to develop computers that would otherwise be impossible or at least extremely impractical. [Sources: 14, 15, 17]
Called Giganto, it is capable of integrating a 12-inch silicon wafer with more than a trillion transistors into a single piece of fabric about the size of a human hair. This has the potential to transform an entire server cluster, which would normally require hundreds of thousands or even millions of microprocessors and thousands of chips, into individual pieces of silicon. It has led to a critical learning process that helps us develop next-generation solutions for a computing-hungry audience. [Sources: 5, 14, 16]
The UCLA Technology Development Group says the silicon chip can be reduced 1,000 times, while the circuit board package shrinks only four times. The silicon chips can also be increased 1,000-fold and reduced 500-fold in power consumption, while the PCB's casings shrink only 4-fold, according to UCLA. [Sources: 3, 11]
Gupta and Iyer are investigating how their silicon composite fabric could influence the size of real systems. The history of networked technology and SOC design shows that the semiconductor industry has traditionally lagged behind in the development of materials that connect chips and their individual components on the chip. But as chip applications from processors to mobile devices have become more complex and more extensive, the need for advanced networking structures has become more pressing. According to Gupta, interconnect tissue technology has evolved to keep pace with modern CPU and GPU technology. [Sources: 15, 18]
Compared to previous crosslinks, Arteris NoC mesh requires fewer connecting wires, simplifies the timing of closures, is easier for chip designers to use and is easy to use. Si - IF has all the properties of connecting a chip directly to silicon, so it is difficult to replace the chip, but necessary to ensure redundancy. Silicon composites consume less power, can switch from circuit board to circuit board at higher speeds without external cables, and have lower power consumption than other materials such as copper. [Sources: 5, 7, 18]
Sources:
[0]: https://www.techbriefs.com
[1]: https://www.engr.colostate.edu/nocs2019/program/
[2]: https://hub.packtpub.com/silicon-interconnect-fabric-replace-printed-circuit-boards-new-ucla-research/
[3]: https://techxplore.com/news/2019-02-gpu-news-waferscale.html
[4]: https://www.nextplatform.com/2019/02/11/giving-waferscale-processors-another-shot/
[5]: https://www.servethehome.com/cerebras-wafer-scale-engine-ai-chip-is-largest-ever/
[6]: https://resources.altium.com/p/pcbs-vs-multichip-modules-chiplets-and-silicon-interconnect-fabric-news
[7]: https://gigazine.net/gsc_news/en/20190929-goodbye-motherboard/
[8]: https://patents.google.com/patent/WO2019089935
[9]: https://www.design-reuse.com/articles/37740/ten-reasons-interconnect-matters.html
[10]: https://anysilicon.com/case-integrating-fpga-fabrics-cpu-architectures/
[11]: https://techtransfer.universityofcalifornia.edu/NCD/29203.html
[12]: https://en.wikipedia.org/wiki/Wafer-scale_integration
[13]: https://www10.edacafe.com/blogs/whatwouldjoedo/2019/03/15/major-stories-this-week-interconnect-fabric-risc-v-ecosystem-arm-neoverse-usb4/
[14]: https://gilderpress.com/2020/01/17/crucial-next-steps-in-the-microchip-revolution/
[15]: https://www.tomshardware.com/news/researchers-kill-motherboard-silicon-interconnect-fabric,40475.html
[16]: https://wccftech.com/amd-coherent-interconnect-fabric-gpus-cpus-apus/
[17]: https://www.techspot.com/news/78544-researchers-developing-better-interconnects-multi-gpu-systems.html
[18]: http://www.arteris.com/blog/bid/99797/The-SoC-Interconnect-Fabric-A-Brief-History
Chiplets on a Silicon-Interconnect Fabric Will Make Computers Smaller and More Powerful
Major companies have long been driving electronics innovation, and the former is evident in everything from computers to mobile phones and mobile phones to computers and tablets.
The latter is defined by today's commercial data centers, whose megawatts devour space for every purpose - and serve monsters in every warehouse built worldwide. The fault lies in the circuit board, and the solution is to get rid of them all and make progress in this arena.
Research shows that the circuit board could be replaced by the same material that makes up the chips attached to it, such as the silicon-silicon compound.
The move would lead to smaller, lighter devices weighing less than wearables and other devices, and incredibly powerful computers that would pack the computing power of dozens of servers onto a silicon wafer plate - a plate. This allows many more chips on a chip connection that can transfer data faster and consume less energy. The wiring between chips and fabric is just as small as the wiring on the chip, except for the connection to the PCB. Silicon technology called silicon-silicon allows the bare chips to connect directly to the wiring without a separate piece of silicon.
Silicon - Composite Fabrics (Si - IF) offer an additional bonus, according to a recent publication in the Journal of the American Chemical Society (ACS).
Instead of SoCs, system developers could develop smaller, simpler chips that connect to more easily manufactured, powerful, and efficient computers. It is an excellent way to solve the problem of producing chips that currently control everything from smartphones to supercomputers.
The chiplet revolution is well advanced, with AMD, Intel, Nvidia and other chiplets offering to assemble advanced packages. A silicon composite fabric expands this vision and breaks the system down into packages, including entire computers. To understand the value of eliminating the PCB, consider what happens in a typical SoC: silicon tape is inserted into a normally plastic packaging that can be up to 20 times the size of the chip itself, while most of us start and end with a PCB.
The size difference between chips and packaging creates at least two problems: First, the volume and weight of the packaged chip are much larger than the original silicon piece. And of course there is the usual argument that everything must be small, thin and light.
If the final hardware requires multiple intercommunicating chips, the distance the signal must travel is increased by more than a factor of 10. When chips exchange a lot of data, distance can be a speed and energy bottleneck, and perhaps the throttle point is a throttle point.
In fact, heat dissipation in computer systems has been a limiting factor for decades. To make matters worse, packaged crisps are hard to keep cool, so if that's a problem, why not just remove the packet?
The purpose of the PCB is, of course, to integrate chips, passive components and other components into a functioning system.
It is not ideal technology, but it is not the only one with potential for use in a wide variety of applications and applications.
In other words, you cannot pack more than 400 ports into a square inch of chip space. PCBs tend to warp and are difficult to flatten perfectly, so the surface warms and warps. The chip housing is connected to the circuit board by solder dents, which are melted and resolized during the manufacturing process. This solder delle cannot be less than 0.5 millimetres apart and can cause deformation of the surface on the chip.
In many applications, there are far too few connections to power the chip, and in such a small space it only provides the power needed to be a powerful computer - efficient, low-performing, high-performance - in the smallest of spaces.
Recently, the semiconductor industry has tried to limit the problem of printed circuit boards by developing a new form of silicon - the interconnect tissue (SICF). To make the connections work in mathematical units, the designers use chip packets. These packages are picked up by a single piece of paper with a width of 1.50 micrometers and fanned out by this into a cardboard box of 500.
The buffer consists of a small number of bare silicon chips mounted on a thin layer of silicon and connected to the packaged chips. This arrangement increases the complexity to solve other problems, but the interposer and its chips still need to be packed and mounted on a circuit board. This would allow a single chip to be produced in two packaged chips - chips totalling 10,000 chips per chip.
RFs can be directly connected to the silicon, but the cache system is thin, fragile and limited in size, making it difficult to build large systems around it. He believes that a better solution is to get rid of the packaging and the circuit board altogether and stick the chips on the chip instead.
Compared to a fiberglass epoxy composite called FR-4, the silicon wafer is stiff and can be polished to flatness, but no cooling is required and distortion is no longer a problem. The solder bumps can replace micrometer columns that are applied to silicon substrates, and the I.O. ports can then be glued directly to the columns.
By carefully optimizing the thermal bonds, compression bonding allows the production of copper-to-copper bonds, which is more efficient than brazed bonds that consume less materials. The elimination of the PCB and its weaknesses means that I.O. ports and chips can only be separated from each other by 10 instead of 500, and packages that need space for transformers can be packaged in one package instead of two or three in one package.
Even better, the standard semiconductor manufacturing process can be used to generate multiple layers of wiring (Si-IF). The technology can achieve chip-to-chip distances of less than 100% compared to a printed circuit board, and the tracks can all be much finer than on a printed circuit board. I.O. ports and chips in a single package of Si-IF chips can be less than 2% apart compared to 500% of a printed circuit board.
The result is a Si-IF system that saves space and power, and shortens the time a signal takes to reach its destination.
Silicon is a relatively good conductor of heat, in addition to the PCB material (chip package), and the processor runs faster by dissipating more heat. IF heat sinks are mounted on the chips to extract more of that heat - some estimates suggest up to 70 percent more when extracted.
Silicon is a bit brittle, but it has very good tensile strength and stiffness, and fortunately, the semiconductor industry has been developing new ways to handle large silicon wafers without breaking them for decades.
However, when SI-based systems are properly anchored and processed, we expect them to meet or exceed most reliability tests, including high speed, low performance and high reliability. The cost of materials for crystalline silicon is higher than FR-4, but we are confident that we are ahead of the curve in terms of performance and reliability, as well as cost. Although many factors contribute to the cost, there is a significant difference in the number of silicon wafers per square centimeter and the size of each silicon chip. However, our analysis shows that if you take out the space - which includes savings - the cost difference is negligible, at least for silicon composite systems.
The silicon PCB size can be reduced from 1,000 cm2 to 400 cm 2, according to the study. A study on server design found that using packageless Si-based processors can double the performance of traditional processors with increased connectivity. Let us look at how the integration of Si computer systems can benefit and how it can reduce computer costs.
On the other hand, there are small Internet of Things, systems based on the arms of microcontrollers. These systems have the necessary cooling infrastructure, but they are shrinking without so much real impact.
By shrinking existing systems and increasing performance, Si-IF will give system developers the opportunity to develop computers that would otherwise be impossible or at least extremely impractical by shrinking them. The plate is 70 percent smaller and its weight has been reduced from 20 grams to 8 grams.
High performance computing applications require multiple servers, and a typical high performance server contains two to four processors and printed circuit boards. Communication latency and bandwidth bottlenecks occur when data is exchanged between different processors on a circuit board.
All these processors are embedded on the same silicon wafer and integrated into a whole system that is larger than the processor.
This concept was first proposed by Gene Amdahl in the early 1990s as part of his research project trilogy. The trilogy failed because the production process could not produce enough functioning systems.