Moore's Law Will Survive Thanks to Silicon Chiplets

20-12-2018 |   |  By Christian Cawley

It used to be a promise, a guarantee; later it became a challenge. But these days, 50 years on, Moore's Law has become a spectre, a shadow threatening to derail processor research.

Of course, Gordon Moore never expected development to increase exponentially, and warnings have been made over the years that things will have to change. But thanks to silicon "chiplets", a new form of processor, Moore's Law looks set to continue for a while yet.

Moore's observation that chip sizes shrink while costs are halved started off as an annual rule which then dropped to 18 months. As of 2016, the rule was suspended, but engineers still aim to stick as close to the 18-month cycle as possible.

Forget Graphene and Qubits

Quantum computing has long been expected to prolong Moore's Law and become the new paradigm in computing. Utilizing quantum mechanics, and with data sent as quantum bits (qubits), quantum computing has the potential to revolutionise encryption.

Unfortunately, while graphene has been identified as the perfect material for quantum computing, external influences on a quantum environment prove dangerous. Heat, motion, sound, and other interactions can destroy qubits, a phenomenon known as decoherence.

This looks set to slow quantum computing development for some time to come.

Chiplets: High-Tech Lego

Taking a modular approach to the design and sale of processors, researchers at Intel, AMD, and the Pentagon believe that chiplets - manufactured pieces of silicon that can be connected - can keep pace with Moore's Law.

By producing chiplets in bulk, dedicated CPUs can be pieced together to meet the requirements of any hardware manufacturer. As well as serving the needs of the end user, this approach also opens a new sales funnel for resellers.

Chiplets Are Already Here

Several chiplet-based CPUs are already available, such as AMD's "Epyc 2" server CPU, comprising eight chiplets. It's first server chip in several years, the Epyc 2 has enabled AMD to increase its share in the server market from 1% to 1.6%.

AMD's lead in the release of chiplet-based hardware - Intel are struggling to release its own - puts them ahead of the game for the first time since the release of 64-bit CPUs back in 2003.

Announced in November 2018, the Epyc 2 CPUs are based on the Zen 2 microarchitecture, combining 7nm and 14nm manufacturing technology on the same chip. All components are designed to the Infinity Fabric standard, AMD's proprietary interconnected architecture, which ensures that data and control is correctly transmitted across all component chiplets.

A Future of Custom Computing

Thanks to chiplets, processors can be created for very specific tasks. Constructing a CPU out of chiplets reduces the length of data connections, potentially cutting major bottlenecks. So, a computer purchased specifically for CAD would have a CPU designed for that task, rather than a multipurpose CPU such as an Intel Core i5.

Similarly, a server streaming video data would have a very different CPU than one intended for serving static web pages.

Thanks to chiplet technology, machines of apparently similar specification could be focused on two completely different purposes. We could be looking at a future where not only are computers custom-designed to handle specific tasks, but so too are the CPUs.


By Christian Cawley

Christian Cawley is a freelance technology writer, with a background in healthcare and financial services industries. He writes extensively online, and contributes to print periodicals and specials.

Related articles