Intel would be finalizing the development of an interconnection system between different elements of the processor DIE with different manufacturing processes, which it has called EMIB.

The AMD Ryzen processors and the AMD Ryzen Threadripper are the first to reach the market with a modular system, although it is not ruled out in the future that this system be implemented in other processors. Intel has already developed other systems, such as the 'Mix & Match' design, which allows the parts of the processor to be separate parts and is made up of optimized manufacturing processing. Intel wants to divide the processor into several smaller arrays, allowing the parts that do not scale well to smaller nodes to be built into mature process nodes, allowing the use of different manufacturing processes.

However, problems arise when the matrices are connected to each other, since this interconnection must be of low power, low latency and high bandwidth to be effective. The problem is the connectivity of the nodes and their density, allowing as a solution, a multi-chip encapsulation system or through the interposition of silicon. The multi-chip would be the one used by AMD today and it is quite simple, although the drawback is that these interconnections must be separated from each other and the number of interconnections is quite limited.

Another solution is the interposition between the matrices and the silicon of a substrate that allows the interconnection of all, which will allow a fast interconnection, while also allowing the silicon to interact with the package underneath the TSV technology. This solution is used for the interconnection of the HBM memories with the GPU arrays in the Fiji and Vega series. This solution has the problem of high costs, which increase more with respect to the necessary size for the interconnection base, making it a disastrous solution for processors with several DIE and large size.

Intel has developed the Embedded Multi-die Interconnect Bridge (EMIB) technology, which would solve some of the problems posed by the previous solutions, making use of silicon bridges integrated into a substrate to act as highways of dense connectivity between the different matrices. This allows silicon the ability to use matrices to pack interconnects in the normal way, without going through a middle ground, custom designed using the largest complex multi-die systems.

Source: OC3D