At their 2018 Architecture Day, Intel revealed their development strategy for future processors - revolving around the concept of breaking down elements from the conventional CPU into singular, stackable ‘chiplets’. Targeted for late 2019, Intel is aiming to offer these products built on what is termed Foveros 3D stacking - the first ever application of stacking processing component inside a chip.
Similar things are being done to storage memory with 3D NAND, and it seems Intel is pushing for the same in the CPU, basically allowing designers to stack extra processing power on top a pre-assembled chip die. So, memory, graphics, power regulators, and AI processing can be in the form of chiplets that can be stacked on one another. More than just providing versatility and better processing, this modular approach allows Intel to circumvent one of their biggest challenges they are facing right now - building complete 10nm chips.
Media reports published today that Intel is ending work on the 10nm process are untrue. We are making good progress on 10nm. Yields are improving consistent with the timeline we shared during our last earnings report.
— Intel News (@intelnews) October 22, 2018
However, Intel has claimed that they are “making good progress on 10nm”, so it isn’t unthinkable for them to implement both stacking 10nm, 14nm and 22nm chiplets on a base chiplet as they’ve suggested with 2D Stacking (seen in image below).
Intel has also promised improved latency and allowing more operations to be running in parallel with codename ‘Sunny Cove’ - essentially acting more like a GPU. ‘Sunny Cove’ will be at the center of Intel’s next generation Core and Xeon processors in the second half of 2019. It is unknown if Foveros 3D stacking will be part of ‘Sunny Cove’ but when asked on the implication for devices, Intel commented that it would range from “mobile devices to data centers”.
Regardless, it seems that bigger things will be coming from Intel in smaller packages in 2019.