The smartphone industry is massive. Each and every year, the revenue from smartphone sales is about US$400 billion and over a billion units are sold each year.
Google, who owns the android ecosystem is taking a page out of Apple’s handbook. They are going to be designing their own in-house smartphone chip.
It will debut in Google’s upcoming Pixel 6 smartphone later this year.
Most Android phones including Google’s own Pixel line have relied on the chip designer Qualcomm. To clarify, Qualcomm an American company, designs the chips but the chips itself are actually built by Samsung and TSMC.
The latest move from Google looks to remove itself from Qualcomm completely. The new in-house processor codenamed Google Silicon 101 or otherwise known as Whitechapel will power Google’s upcoming phones. It is also expected to feature in future Chromebook.
Samsung has teamed up with Google to build their custom chip and this makes sense as they didn’t only build chips for Qualcomm but also use their own custom chips in their smartphone sold outside the United States. These Samsung chips are known as the Exynos series.
Google CEO Sunder Pichai stated that Google planned to seed more investment into hardware but as it stands, Google hasn’t invested enough in their semiconductor business to be competitive just yet although they are choosing to hone their focus in an area where they can obtain a competitive advantage against the major players.
And this area is Artificial Intelligence (AI). Google is already good at using AI to enhance image quality; the camera hardware on their phones hasn’t been updated in years but manages to keep up with competition solely thanks to the AI within their software.
A chip with a focus on AI hardware acceleration could lead to even better photos. Using machine learning, improved color signs, image stabilization, and even audio processing is all possible.
A report from Axios mentions the Whitechapel chip will have a dedicated area for Google assistant processing. This is to improve performance and always-on capabilities.
With Google’s Lamda coming on the software side, this could be interesting. For those of you who don’t know, Google Lamda is going to incorporate the latest in the AI language processing, so the Google assistant can understand context and have conversations with its users.
Lamda plus custom hardware could make for a next level AI assistant potentially even offline.
One of the biggest advantages of building an in-house chip over buying an external one off the shelf is the tight integration between software and hardware. The software knows exactly what it is going to be working with so it can be more efficient. This should improve the end user experience.
In the earlier day of smartphones, heavy optimization from Apple controlling both the software and hardware allowed the iPhone to operate smoothly and provide an arguably better user experience despite being woefully underpowered and limited in features compared to the competition of the time.
Another benefit of Google building their own chips is a reduction in production cost. Now, instead of acquiring processors from external providers like Qualcomm, who themselves need to make a profit, Google can design the chip themselves. And if they choose, they can use the money saved to either cut down costs or use the money to improve other areas of the phone.
Google’s decision to separate from Qualcomm is an interesting one because it is actually going to look like there’s going to be a performance drop when compared to the best that Qualcomm has to offer.
The new Whitechapel chip is going to be behind Apple’s a14 bionic, the snapdragon 888 and definitely not to mention the M1.
According to Wccftech, we know from the internal documents and other leaks that the Whitechapel chip will have an 8-core-arm design with three clusters. Much like the Samsung Exynos 2100. And while efficient compared to the company’s previous manufacturing techniques, the Google chip will most certainly be a step behind the competition.
9to5Google states that the predicted performance will be similar to the Qualcomm snapdragon 765, already used in the Pixel 5. To the layman, this just means that the Google chip will have middle of the road performance.
Interestingly, according to the Economic Times, a team of Google researchers are working on an AI model that allows chip design to be performed by artificial intelligence.
The new AI method utilizes past experience to become better and faster at solving new instances of the problem. The Google team wrote in the paper for the scientific journal nature, “Our method was used to design the next generation of Google’s artificial intelligence accelerators, and has the potential to save thousands of hours of human effort for each new generation.”
According to the team, in about six hours the model could generate a design that optimizes the placement of different components on chip, “In under six hours, our method automatically generates chip floor plans that are superior or comparable to those produced by humans in all key metrics, including power consumption, performance and chip area.”
To achieve this, the Google team used a data set of 10,000 chip layouts for a machine learning model which was then trained with reinforcement learning.
Anna Goldie, a research scientist at Google Brain explains, “Our reinforcement learning agent generates chip layouts in just a few hours, whereas human experts can take months.”
Despite many decades and a mature industry, chip designs has resisted automation, requiring months of intense effort by physical design engineers to produce manufacturable layouts.
So this latest development is extremely interesting. We are going to have to wait and see if Google will make up for lost time by using AI in their upcoming chip designs. It is also important to note that with the global semiconductor chip shortage, Google might have a tough time actually getting this chips built.