The chiplet paradigm and what it means for the future of silicon architectural design?
Chiplets are a radical shift in the way we design and manufacture microprocessors. The small, modular and cost-efficient pieces of silicon can be combined to create custom microprocessors that are adapted to market requirements, whilst offering a wide range of flexibility and customisation in the design process. This story aims to explore what chiplets are, explain their strengths and compare them to more traditional(monolithic) approaches to microprocessor design.
What is a monolithic microprocessor design?
Monolithic designs have mostly been used to create microprocessors up until very recently, this involves the entire chip being created from one large piece of silicon containing all the components of the die including transistors, circuits and other specialised hardware.
The monolithic approach has been the main method of creating microprocessors since the release of the Intel 4004 in 1971 and has a range of advantages such as being simple to manufacture and being more reliable since all the components are integrated into a single chip.
What is the chiplet paradigm?
The consumer-focused chiplet paradigm was pioneered by AMD in 2019 when the ‘Zen 2’ microarchitecture was released. It contained a new approach to silicon architecture design, which involved using small modular pieces of silicon that are connected together using an interposer. These ‘chiplets’ can then be designed and manufactured separately to form larger and more complex microprocessors.
The modular nature of chiplets allows greater customisation and flexibility in the design process of a CPU and allows semiconductor companies to segment their products more efficiently by adding or removing the number of chiplets. Another advantage to using chiplets is the reduced manufacturing costs, since chiplets can be manufactured in different facilities, it allows the use of multiple manufacturing processes and technologies depending on the application of a chiplet. This can also increase efficiency as not all hardware required on a chip needs to be on the latest manufacturing technologies and thus are contained on an IO die as seen below which often use older and more cost-effective manufacturing technologies.
What chiplets mean for the future of Silicon design?
The chiplet paradigm involving the use of small modular pieces of silicon represents a significant shift in the way microprocessors are designed and manufactured. As they allow for greater flexibility and customisation as well as the ability to optimise performance due to their scalability. These benefits are driving the adoption of chiplets in microprocessors and can be seen in the latest architectures from semiconductor companies such as AMD, Nvidia and Intel.
