PCBs vs. Multichip Modules, Chiplets, and Silicon Interconnect Fabric
2019 saw a resurgence in talk of multichip modules, chiplets, and silicon interconnect fabric. An article in the September issue of IEEE Spectrum claimed that silicon interconnect fabric, a method for connecting chiplets on a multichip module, would eliminate PCBs and bulky SoCs for many applications, specifically motherboards. Will any of this ever come to pass? Obviously, we can’t know the future, but designers should be aware of changes in the electronics industry.
Although this is at least the third time this has happened in as many decades, multichip modules date back to IBM’s Bubble memory in the 1970s. Once you cut through the buzzwords and analyze the challenges involved in bringing multichip modules into the mainstream, it’s easier to see what the future relationship between PCBs and integrated circuits looks like.
Multichip Modules, Chiplets, and Silicon Interconnect Fabric
If you’re not familiar with these three terms, these technologies are interrelated parts of heterogeneous integration. In this type of integration, different devices with individually optimized structures are integrated into a single wafer.
- Multichip module: Think of this as an entire PCB, but made out of silicon or another semiconductor, with everything built into a single wafer.
- Chiplet: This is effectively an integrated circuit block on silicon or another material. This block is built directly onto a multichip module using a lithography process.
- Silicon interconnect fabric: The underlying interconnect architecture that connects chiplets into a multichip module. Think of this as the silicon analogue of traces on a PCB. Note that the same interconnect architecture can be adapted to other semiconductor materials.
Heterogeneous Integration Challenges in Multichip Modules
Multichip modules are something of a dream technology in the electronics community. The benefits center around elimination of plastic packaging, and on reduction or elimination of external interconnects. With this technology, the vision is to eliminate multiple SoCs and the traditional SiP structures with an entire system built onto a single wafer. If you’re building out SoCs for smartphones, this would effectively integrate multiple ICs with different functions into a single multichip module. This is the essence of homogeneous integration.
The challenge of this approach is that heterogeneous integration is not really heterogeneous. It is only heterogeneous in terms of functionality, but not in terms of materials and proceses. Different ICs with different functions can be heterogeneously integrated as long as they are developed with the same materials and processes. While you can certainly integrate features like power regulation, RF transceiver functions, cores, memory, and other typical ICs onto a single wafer, sacrifices need to be made in terms of optimization for different functions.
RF amplifiers for microwave/mmWave applications are going the route of GaAs, GaN, and SiC materials, while memory and processing functions are still largely confined to Si. However, GaN is one material that is gaining ground (e.g., releases of GaN microcontrollers) and may provide an opportunity for heterogeneous integration of power circuitry, GHz RF circuitry, memory and processing power into a single wafer.
Given the technical challenges involved in integrating disparate chiplets with different functionalities and materials on a single multichip module, PCBs are here to stay. The problems in reliably integrating multiple chiplets with different functions is currently prohibitive, although this may change in the future. This means different functions requiring different materials would need to be separated into different multichip modules. These separate multichip modules would then need to be integrated into a PCB just like any other group of ICs using standard assembly processes.
Until every function can be integrated into a single wafer, PCB designers will still have jobs designing advanced electronics systems. In this researcher’s opinion, we’ll see electronic-photonic integrated circuits (EPICs) become heavily commercialized before we see the heterogeneous integration envisioned in multichip modules. We may even see photonic chiplets integrated into multichip modules and connected with a photonic analogue of silicon interconnect fabric. Indeed, companies like Vanguard Photonics have been working on this technology for some time, and the integrated circuit industry is holding conferences on developing standards and scaling strategies for commercializing silicon photonics components.
Even if we do see multichip modules with significant heterogeneous integration, this doesn’t mean we’ll have complete elimination of PCs. With EPICs, a PCB will still be needed to provide optical interconnects between components. With photonic multichip modules, there are likely to still be problems with integration due to incompatibility between Group IV, III-V, and II-VI semiconductors. The wealth of electronics produced every year for the consumer market is highly diverse, and not all of these products will need something as advanced as multichip modules.
As electronics continues to advance into areas like multichip modules, look to companies like Altium and design platforms like Altium Designer® for tools to work with these new technologies in PCBs. The toolset in Altium Designer sits at the cutting edge of the EDA industry. You’ll be able to design compact boards for any application, manage your design data, and prepare new products for full-scale manufacturing.