3

A Changing Industry: The Rise of In-house Silicon Picks Up Steam in 2021

 2 years ago
source link: https://www.allaboutcircuits.com/news/a-changing-industry-the-rise-of-in-house-silicon-picks-up-steam-in-2021/
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.

News

A Changing Industry: The Rise of In-house Silicon Picks Up Steam in 2021

December 02, 2021 by Jake Hertz

A shift to in-house silicon is redefining the dynamics of the industry and shaking up long-standing "norms;" however, why is this happening and starting to gain momentum with larger companies?

Over the past couple of decades, the dynamic of the semiconductor industry has been well defined and unchanged. Only a handful of companies, namely Intel, AMD, and Qualcomm, were involved in processor design work, while the rest were simply consumers. 

bulletin0525Fig01.jpg

The top 15 semiconductor companies by sales numbers, which also show Apple as a contender with in-house chips at number 10. Image used courtesy of IC Insights

This scheme went largely unchallenged for many years until it seemingly took a complete 180 in recent years. Today, in 2021, tech giants, including Apple and Google, are starting to take silicon design into their own hands. 

This article will explore this trend, why it might be happening, and what it could mean for the semiconductor industry's future. 

Silicon Trends in 2021 

The past 12 months have seen an enormous amount of change in the semiconductor industry, most notably with the release of in-house system-on-chips (SoCs) from Apple and Google. 

While Apple has been creating its own processors for the iPhone for years, it has exclusively used Intel processors in its Macintosh computers. That changed late last year with the release of Apple's M1 SoC, the company's first in-house chip for Mac. 

The company doubled down on this approach more recently when they expanded their portfolio by introducing the M1 Pro and M1 Max.

Three SoCs, two generations, all 5-nm node size.

Three SoCs, two generations, all 5-nm node size. Image used courtesy of Apple 

Google followed suit shortly after, releasing Tensor last week, its first custom in-house SoC. 

Tensor represents a shift for Google, which has historically used Qualcomm's Snapdragon family processors in its Pixel devices. 

The trend does not stop here, though. Other companies, including Alibaba Cloud, have taken to creating their own silicon. Reports even indicate that major players like Facebook are investing in the same. 

With all of these companies making this internal switch, the question that comes to mind is why?

Why Go In-House? 

The major reason that most of these companies cite has to do with the end-to-end control of their products. 

Even though Apple makes Apple products and Google makes Google products, when using someone else's silicon, these products are confined to the capabilities of that silicon. 

That is to say, Apple and Google may have visions of what they want their products to do, and as long as they're restricted to Intel or Qualcomm's processors, they'll never be able to achieve them. 

Notably present in both companies' SoCs are artificial intelligence (AI) accelerators, hardware blocks meant to improve performance and reduce power consumption for AI applications explicitly. These accelerators could provide another hint as to why companies are bringing silicon in-house. 

Historically, computing tasks have been general-purpose; standard CPUs with ubiquitous architectures were relied on to take care of most tasks. Fast forward to 2021, and AI and machine learning (ML) pose unique computing requirements. Not only is hardware that can specialize in AI/ML workflows needed, but also hardware that is specialized to employ specific ML models is also necessary. 

Hardware that can accelerate a convolutional neural network (CNN) may not be effective at accelerating a different AI task.

An example of an ML accelerator.
An example of a CNN accelerator. Image used courtesy of Hua et al

Apple and Google may both have voice assistants within their devices; however, the underlying ML models and architectures that they use could differ dramatically. 

As a result, making in-house silicon allows companies like Apple and Google to fine-tune their products for their specific software applications and architectures. In this way, everything is perfectly cohesive and proprietary, from the software to the hardware. 

A Custom Silicon Future? 

When major tech players take a dramatically different approach to the industry standard, it usually is indicative of a paradigm shift to come. In the coming years, one can expect to see more and more major players turn to in-house silicon, while companies like Intel, AMD, and Qualcomm will need to search for new ways to turn profits.


About Joyk


Aggregate valuable and interesting links.
Joyk means Joy of geeK