– and why it hasn’t happened yet
Chiplets are hot! According to all the buzz, chiplets are poised to disrupt just about every sector of the semiconductor industry. What’s behind all the excitement? Let’s talk about that. But first, a definition:
Chiplet: A small modular integrated circuit, containing a well-defined subset of functionality, designed to be combined with other chiplets in a single package to create a complex component.
So, are these chiplets going to take over the world?
The system on chip (SoC) solution is firmly established, with big all-in-one chips controlling everything from smartwatches to personal computers. In the chiplet vision, however, an SoC is replaced by a collection of chiplets integrated via advanced packaging. Can this new technology upend the reigning paradigm?
It all comes down to money, money, product improvement, and business advantage (i.e., money).
1. Reducing the Cost of Manufacturing
Chiplets offer several ways to make manufacturing less expensive. The most obvious is the reduced cost of silicon area. Chiplets are tiny. Smaller dies pattern a wafer more efficiently, leaving less wasted silicon around the edges. Smaller dies also yield better: a flaw that would destroy a large chip will instead destroy just a little one, and a greater percentage of the wafer’s silicon survives.
Another cost-saving advantage is “known good die” (KGD) testing. Manufacturers test components before assembly so that one bad piece doesn’t doom the entire device. One bad spot in an SoC ruins the whole SoC; one bad chiplet can be detected and discarded before assembly, preserving all the other components.
A final cost factor is less obvious: mixed nodes. Chiplets built at different nodes can work together beautifully. Analog functions – such as power converters and sensors – don’t need the latest bleeding-edge technology node. They actually work much better in larger geometries. These mature geometries are both more stable and significantly less expensive to process. Larger, old-node analog chiplets save enough in processing costs to more than offset the extra silicon area.
Taken together, these advantages make manufacturing a set of chiplets more cost-effective than producing an equivalent SoC.
2. Reducing the Cost of Design
Designing state-of-the-art microelectronics is an art just short of wizardry. It’s deeply complex, exciting, and flat-out expensive. Chiplets can help with this cost, too.
Time is money. A chiplet-based system allows design engineers to work in parallel, crafting their circuitry within cleanly defined boundaries. Eliminating overlap and “fuzzy edges” delivers a finished system more quickly.
Engineers often shorten the design cycle by purchasing and incorporating pieces of existing circuit designs (intellectual property, or IP). Chiplets address the IP issue in two ways: First, chiplet designs rarely require purchased IP because chiplets are meant to be small and simple. Second, a chiplet itself is the epitome of reuseable IP – a well-defined subset of functionality in hardware form. Tidy, well-executed chiplets can be re-used in multiple products, saving time and money in each design.
Chiplets really come into their own during redesigns. With a chiplet-based system, only specific chiplets need to be updated, leaving the rest of the system intact. Design and test are greatly simplified; schedule, cost, and associated risks are dramatically smaller.
Mixed nodes also reduce design cost. As mentioned earlier, some chiplet functions work better at bigger nodes, which brings another excellent benefit: bigger nodes make design work a lot easier. Shorter schedule; cheaper design.
All in all, devices assembled from chiplets are less expensive to design (and re-design) than SoCs.
3. Improving System Performance
Money is a strong argument, but an even bigger incentive to use chiplets is performance. A single chip is necessarily built on one material, at one node, using one process flow. When that chip is a multi-function SoC, compromises are inevitable.
Because chiplets are manufactured separately, they can take advantage of the full range of materials, nodes, processes, and even sources. When assembled with advanced packaging, the potential for optimization is profound.
Nodes
As already noted, different functions excel at different nodes. Specifically, small geometries make faster transistors and are therefore ideal for digital logic, while large geometries build better analog devices. Larger nodes are also more rugged, can handle higher voltage and current, and suffer less noise. Fabricating each chiplet at its optimal node creates best-in-class functionality. No compromises.
Processes
Even at the same node, wafer processes are modified to optimize different elements. Logic processes produce fast transistors; memory processes minimize leakage in memory cells; photonic processes favor modulators and waveguides; RF processes optimize inductors, caps, and resistors. Manufacturers can select the best process, chiplet by chiplet, to get top performance.
Materials
Silicon is king. Nothing supports high-density logic and memory like silicon. However, other wafer substrates out-perform silicon in specific niches:
-
- SiC and GaN are superb for building power converters.
- InGaAs and GaSb make the best infrared detectors.
- InP rules the radio-frequency realm.
These materials are seriously pricey and none can be used for high-density logic or memory. They can, however, be employed in a system as tiny, powerful, function-specific chiplets.
Sources
This is still just a gleam in the designer’s eye, but think about it. When leading chipmakers start selling commercial chiplets, system designers will pick and choose premium components from each of them to create top-notch systems. The very best memory; the very best sensors; the very best processors, heat spreaders, transmitters. Wow.
With so much scope for optimization, chiplets are slick, keen, potent little engines. Advanced packaging supports and maintains their high performance, crafting them into extraordinary new systems. Chiplets will unleash a revolution in flexible system design.
4. Increasing Product Appeal
Customization is a huge business advantage. A stripped-down economy car appeals to one buyer; a flashy power beast attracts another. Car manufacturers learned long ago to satisfy a whole spectrum of buyers by making base models with wide arrays of optional features.
Chiplets empower similar diversification in electronic devices. Rather than creating each model from scratch, manufacturers can use chiplets to broaden their product lines. Plugging different chiplets into a base design can adjust the number of processor cores, swap in specialized interfaces, or customize the type and amount of onboard memory. Hot new technologies can be implemented at a fraction of the cost of a whole new system, and in a fraction of the time.
Custom designs – or even semi-custom – at affordable prices? An awesome business opportunity for forward-thinking manufacturers.
What’s Holding Back the Chiplet Revolution?
If chiplets are destined to take over the world, why isn’t everyone leaping aboard the band wagon right now? Well, it isn’t quite as easy as it sounds. Chiplets don’t just pop into place like LEGO® blocks.
Multi-die assembly is complicated. Bonding and packaging approaches abound, each with advantages and drawbacks. Varying die area, height, bump pitch, material, and thermal qualities must be taken into account.
And then there’s testing. Even with KGD, the final assembly must be tested as a whole. It takes a sophisticated, multi-faceted process to ensure that all the various dies and their connections are sound.
The supply chain is a whole other can of worms. Chiplets may come from many suppliers, greatly expanding the sources to be managed. Each source introduces its own set of definitions and risks; each requires active communication; each presents a potential disruption in the chain.
No, the chiplet solution isn’t a walk in the park. But it can be done. Eager teams of technical professionals are already hard at work, making it happen.
The Future is Chiplets
Yes, chiplets will take over the world! It’s as inevitable as a Chicago winter. Assembly and testing challenges will be overcome; supply chains will be tamed; unforeseen problems will arise and be conquered. The implacable force driving this transformation? Economics.
Chiplets assembled with advanced packaging are the new system-level realization of Moore’s Law. In his 1965 paper, Gordon Moore correctly projected radical increases in functionality and transistor count with minimal increases in cost. For decades, this model has propelled the stratospheric growth of the semiconductor industry.
But on page 3 of that same famous paper Moore states: “It may prove to be more economical to build large systems out of smaller functions, which are separately packaged and interconnected.” The industry is now turning to page 3. Gordon Moore nailed it again.
Semiconductor technology is inventing the future – one brilliant chiplet at a time.