The basis of Moore’s Law is simple: the number of transistors in a dense integrated circuit doubles about every two years. Basically, this states that every two years chips will get smaller, have improved power consumption and performance. However, the time period detailed by Moore’s Law is lengthening as chipmakers struggle to make innovations at the same pace. CPUs have become so small that significantly reducing them further is becoming challenging. Chipmakers are increasingly looking for technology that will allow the historic standards of Moore’s Law to be maintained. One tried method is chiplets, while is when a single processing unit is built on multiple chiplets that are connected together.
Chiplet Approach
One of the core benefits of chiplets is they allow manufacturers to more efficiently build CPU components. AMD is one of the several industry heavyweights who are increasingly seeing chiplets as a viable path to continued innovation. In a piece for TechRadar, Mark Papermaster CTO and Executive Vice President at AMD, highlighted how chiplets are beneficial:
“The product needs to have a huge appetite for throughput performance such that the cost of multiple smaller die in the package are significantly lower than legacy monolithic designs, There needs to be a substantial component of analog / mixed signal IP that does not benefit from leading edge technology, The product should benefit from the flexibility enabled by varying the chiplet count across the product line. For instance, at AMD, we sell one, two and four die versions of our “Zen” architecture in AMD Ryzen™, Ryzen™ Threadripper™ and EPYC processors.”
