Systems Of Packages

How and why the concept of subsystems is changing.


The shift from IP blocks to subsystems, with pre-integrated and pre-verified IP, has never lived up to the initial hype. That doesn’t mean the concept isn’t valid, though.

The problem with subsystems, as originally conceived, is they were far too limited for widespread adoption. When this idea first began surfacing in the early pat of the Millennium, this all made sense because the number of semiconductor vendors buying IP had not yet consolidated and Moore’s Law scaling was only rarely questioned. Once market consolidation set in, end users began customizing “standard” IP to differentiate their products from their competitors.

The two subsystems that have proved successful over the years—audio and video—both were developed around the premise that they could be customized to incorporate new protocols and to manage power/performance needs of an end product. These were devised for markets that continue to be in almost perpetual flux, where fixed IP has a very short shelf life.

Elsewhere, there appears to be a growing recognition that the definition of subsystem is too narrow. Rather than pre-integrated blocks, the real opportunity may include pre-integrated systems in packages (SiPs), or systems of SiPs.

This is already happening in the automotive market, where the biggest problem is how to manage all of the data collected from assisted and autonomous vehicles, and in a number of IoT segments, where time-to-market is more important than customization.

Automotive designs for assisted and autonomous driving increasingly rely on systems of sensors to detect the physical environment. Those systems include cameras, thermal imaging, radar, LiDAR, and a collection of inertial sensors such as gyroscopes. Streaming data from all of those devices to a central processing system isn’t a workable solution. There is way too much data to process quickly enough in one location, and doing so would force assisted and autonomous vehicles to slow to a crawl in order to remain safe.

The current thinking is that sensors will be packaged together with entire pre-processing systems—various processor elements, memory, I/O, and most likely some sort of thermal management and error detection and correction. This is much more than a subsystem as it was originally defined, and it points to a much greater emphasis on rugged advanced packaging designed specifically for extreme temperature swings, extended vibration, and every type of noise imaginable. Moreover, because this is still a nascent market, there will need to be enough flexibility built into these packages to adapt to improvements and updates, whether those are over-the-air or downloaded at the dealership.

A number of IoT segments likewise will require those kinds of complex building blocks, but for different reasons. In the IoT, market windows are measured in periods of less than six months, and sometimes less than three months. The only way to get a product from design to market in that timeframe is with packaged components. The real edge in some of these markets is more about getting new products out the door in different market slices than in creating something that is completely unique. So for device makers, the differentiation is much greater on the business side than in the technology. For chipmakers and packaging houses, the differentiation is creating customizable packages of pre-integrated components that can be easily adapted both within and between markets.

While these are variations on what has often been described as bigger LEGOs, these SiPs and systems of SiPs will be much more complex and flexible than the original idea behind subsystems. Time to market and pre-verification and test are still critical elements, but the opportunity and challenges associated with this approach will likely be much more complex than anyone would have imagined.