Chiplets Are Revolutionizing Semiconductor Manufacturing with More Flexibility and Lower Costs

Debate around Moore’s Law and developments from chipmakers at the leading edge of innovation suggest the decades-old adage might be dead. With chips growing increasingly smaller and more complex, doubling transistor counts every two years just isn’t feasible. Yet, buyers demand more powerful components for their data centers, laptops, and cars.  

The solution? Chiplets. Over the last several years, chipmakers have experimented with a revolutionary new way to design and produce semiconductors. Rather than jamming all of their features onto a single piece of silicon, they create multiple chips and package them together to work as a single system.  

AMD and Intel currently lead the way in chiplet research and product offerings. Given this technology’s potential to reshape the entire industry, other chipmakers are now starting to take notice.  

Chip Design of the Future

The idea behind chiplets boils down to a modular design that lets numerous small components do the job of a much larger one. In a chiplet, each small chip makes up some part of the system. For instance, when AMD introduced its Zen processors in 2017, its design featured multiple CPU chips to boost the core count.  

As chiplet design advanced over the years, chipmakers began separating the pieces with distinct functions. Intel’s approach uses modular components called tiles. Each tile has a unique role. The firm’s Ponte Vecchio data center GPU features separate tiles for graphics, caching, memory, and connectivity. The Meteor Lake laptop processor uses a four-tile design while its Arrow Lake successor will bring a similar approach to desktops next year.  

Designing chiplets in this way gives chipmakers more flexibility and control, making it possible to tailor the product to specific markets or use cases. For instance, a chiplet can be designed with more cores or swap some of its chips to deliver additional features like connectivity or caching based on the tasks it will carry out.  

Additionally, the modular design of chiplets makes upgrade cycles both faster and simpler. When a chip or tile needs to be replaced, the chipmaker can do so without changing the other components. Since not every component needs to be upgraded with each generation, this saves time during the design process and makes it possible to offer significant upgrades more often than once every year or two as most manufacturers currently do.  

Easier Manufacturing

Of course, chip design isn’t the only area that benefits from chiplets. Manufacturing is also an important consideration. Indeed, chiplets make manufacturing advanced chips cheaper thanks to their smaller size making them far less prone to defects.  

Production at next-gen nodes, which is particularly vulnerable to high defect rates early in a given lifecycle, benefits even more. Making large chips commercially viable at this stage is notoriously difficult due to increased production costs and limited capacity. By taking advantage of the smaller size of chips integrated into a chiplet, manufacturers can get advanced node components to market faster with lower costs.  

Moreover, the modular design of chiplets makes it possible to produce powerful components that are both smaller and cheaper than they would be as a single unit. Adding dedicated cache chips means CPUs or GPUs don’t need as much cache capacity. The same goes for features like connectivity and memory. In turn, the CPU or GPU can be simpler and dedicated to a single task. Chipmakers can also utilize older nodes for certain components—which are cheaper to produce—without losing performance.  

Ultimately, chiplets won’t overthrow traditional semiconductors for every application. For tiny smartphone chipsets or very simple chips, the design doesn’t make sense. But for the high-powered chips running data centers, smart cars, and consumer devices, the advantage of chiplets is impossible to ignore.  

Over the coming years, expect more chipmakers to follow this trend and pursue chiplet design. Intel and AMD have a strong lead thanks to their early adoption. However, with more attention going toward chiplets, and more funding being put toward design research, the industry could look very different not long from now.