For many years, the monolithic core design was the most popular and widely used option in the semiconductor world, but the increasing complexity of this type of design and the high cost of transfer to the wafer made it necessary to look for other alternatives. more efficient and profitable. AMD found what it needed in the MCM designs and it evolved into a chiplet design.

A chiplet is a small chip that can contain various elements in its own package and is designed to be able to do so be easy to integrate and connect with other chiplets, either by 2D layout, which is the most common, or 3D, which is much less common, although these types of implementations have been shown to have great potential, we just have to see what AMD has achieved with Ryzen with a 3D stacked cache.

Chiplets can contain, among other things, basic parts of the CPU, GPU, I/O elements and also other L3 cache. This greatly simplifies the creation of complex solutions such as a CPU with a high number of cores, because instead of creating a huge silicon chip with a large number of elements we can divide into small interconnected chiplets.

What advantages does a chiplet offer?

I already gave you a preview in the previous paragraph. The chiplet makes it possible to simplify the design and creation of highly complex processors and graphics cores by dividing their parts into several chiplets. Also facilitates on-wafer execution, increases success rateit significantly reduces costs and increases the technical and economic feasibility of designs that would be practically impossible with a monolithic model.

I will give you an example so that you can understand it better. to think about 128 core processorTo make this a reality using a monolithic core design, we would have to have a design and node capable of cramming 128 cores with their caches and all their elements into a single silicon chip. But that’s not all, every chip would have to turn out perfectly, we wouldn’t have room for mistakes.

If a chip comes out with 120 functional cores he would no longer serve usand we would have to dispose of that chip or reuse it to sell it as an inferior product. We could also encounter system I/O or cache failures that could end up killing the chip completely. Since this is a very complex chip, the chances of something going wrong are very high, which reduces its viability on the wafer, increases the complexity and footprint of the wafer, and can skyrocket costs.

We see the CPU chiplet in red and the I/O chiplet in green.

With a chiplet, the approach would be very different, and I’ll use AMD as an example because they were the company that decided the most on this type of design. The Sunnyvale Giant would use it two kinds of chiplets to create this 128 core processor:

• A CPU chip that has 8 cores and 16 threads and has 32 MB of L3 cache.
• An I/O chiplet that integrates the entire connectivity subsystem and memory controllers.

To give life to the 128-core AMD processor you need 16 CPU chiplets and one I/O chiplet. Manufacturing 8-core chips is much easier, the chance of something going wrong is minimal, and the impact is both wafer space and cost It is much lower. But that’s not all, these types of designs are also highly scalable and allow the use of silicon to be fine-tuned to fit the real needs of each user profile and avoid unnecessary costs.

There is also a chiplet highly scalable. Thanks to this modular design, AMD can create very powerful mainstream consumer processors with one or two chiplets (up to 16 cores and 32 threads) and scale of the same design in its AMD EPYC Genoa processors with up to 16 chips that add up to 128 cores and 256 threads.

This type of design makes building processors with higher core counts almost as easy as a Lego game, just add as many CPU chiplets as you need, wire them up, and that’s it.

Chip scaling usually occurs through 2D layout and interconnects, but AMD was also able to scale through 3D stacking, which greatly increased the amount of available cache. The most popular example is in the Ryzen 5000X3D and Ryzen 7000X3D, which have up to 128 MB of L3 cache, but AMD also brought this technology to its EPYC processors and managed to create a solution with up to 1152MB of L3 cache.

A look into the future

AMD’s move to an MCM design with Zen and its commitment to the chiplet with Zen 2 marked an important shift in the industry that was established with Zen 3an architecture that established the chiplet in a sector where the monolithic core design was still trying to present itself as “the only viable option”, when in fact it was starting to become quite the opposite.

AMD’s present and future is passing through the chiplet, but the race in favor of these types of designs It did no more than begin. Its use is gradually expanding and it is only a matter of time when we start to see it in a greater number of technological solutions, from processors to graphics cards, accelerators and other types of powerful computing units.

With the birth of the Universal Chiplet Interconnect Express, the importance and transformational role of the chip was not only confirmed, but also a proposal as ambitious and interesting in the future of this type of proposal was clearly shown, the possibility that the big players in the sector can combine chiplets from different brands in a single package. For example, think of a processor with two AMD CPU chiplets and a GPU chiplet from another brand. Interesting, isn’t it? This would not be possible with a monolithic core design.

Of course, we will also see a solution more efficient and effective which will also have higher cost and viability at the wafer level due to the differential value offered by the chiplet.