Is Advanced Packaging The Next SoC?

Device scaling appears to be possible down to 1.2nm, and maybe even beyond that. What isn’t obvious is when scaling will reach that node, how many companies will actually use it, or even what chips will look like when foundries actually start turning out these devices using multi-patterning with high-NA EUV and dielectrics with single-digit numbers of atoms.

There are two big changes playing out across the semiconductor industry right now that ultimately will have a big impact on the answers to those questions. One involves a growing number of new markets and opportunities around artificial intelligence, whether that’s AI itself or related subcategories such as deep learning or deep neural networks or machine learning.

There are two phases to AI, DL and ML-the training and the inferencing. Training can happen inside a data center using standard hardware, whether that’s GPUs or ASICs. Inferencing needs to happen much closer to the edge, and it requires massive compute power in applications such as automotive because they have to be done almost in real-time. This will require blazing fast processing, but it also will require flexibility because AI and its subcategories today are so inaccurate that they need to be constantly modified.

There are a number of ways to achieve that flexibility. One involves programmability, whether that is in software (easier but slower) or hardware (eFPGAs, FPGAs and PLDs), which is faster but still not as fast as an ASIC. The second involves packaging, where the majority of a device is pre-designed and only a few of the components are swapped out to deal with future needs or protocol updates.

The second big change that is underway is the splintering of markets under the umbrella of the IoT. The growth in cloud computing (regardless of whether that involves internal or external clouds) has fostered much more differentiation at the edge. This is what enables the umbrella terms IoT and industrial IoT. It’s now possible to use a more standardized set of resources on the back end of an operation and customize the front piece, whether that is a point of sale system for a retail store or a system to monitor the amount of liquid or gas flowing through a valve-or even to autonomously plow a field using a GPS-based driverless tractor.

The big impact of splintering markets is economic. Volumes of chips developed for vertical markets pale in comparison to those in horizontal markets, such as communications, which is why smart phones remain the biggest market for advanced ASIC designs. With more than 1 billion devices sold every year, the majority of those by two companies, it’s not hard to see why a decision by Apple to swap foundries or change processors can upset the balance of power in the semiconductor industry.

That could change significantly in the future, though, if more IP is hardened into “standardized” chips that can be used across a multitude of end products. So rather than worrying about developing a 3nm processor, companies can buy one from one or more companies and add it into a package. And because these chips will be used across a multitude of end products, they will have much better characterization and use modeling, which will help with floor-planning and predictability about reliability and aging effects. It still may be an expensive part of the overall design, but that’s far less expensive than building a 3nm SoC where the return on investment would require sales in the order of 3 to 5 billion units.

The era of advanced packaging has begun, regardless of whether some companies are willing to say so publicly. How the pieces ultimately go together, and whether this ultimately re-aggregates or disaggregates the industry isn’t entirely clear. But what will define chips in the future is more about the total package and the optimization of the individual parts rather than the characteristics of the parts themselves. And what will define successful electronics companies is their ability to recognize these changes and build products that can be easily customized for a growing number of end markets.

Related Stories
Toward High-End Fan-Outs
Denser interconnects, stacked die could rival 2.5D approaches.
Fan-Out Wars Begin
The number of low-density packaging options is increasing as the popularity of advanced packaging grows.
Cheaper Packaging Options Ahead
Low-cost alternatives to interposers could have a big impact on chip design.
Packaging Challenges For 2018
Shortages, pricing pressures, rising investments and more packaging options add up to an interesting year for OSATs.