Google and Chips

There have been rumors recently that Google plans to design its own chips. A chip formally known as a SoC is the heart of a smartphone. The SoC a device consists can make or break a device, ask that to anyone who has used a device running Qualcomm's Snapdragon 810.

When it comes to Android Qualcomm, Samsung and Mediatek have been the major SoC suppliers with Intel, Spreadtrum, Broadcomm and several others having minute market share in comparison.

Although Google has some form of hardware presence in Android smartphones in the form of Nexus and Android One they haven't gone down to component level. Building their own SoC would mean that Google would now be making an essential component of smartphones all by their own. Android One and Nexus haven't been run away hits for Google at least in terms of market share, its therefore quite surprising that Google has decided to make an SoC of its own and make its reach in hardware even deeper.

If Google does indeed decide to make an SoC of its own there are significant challenges that the  search giant needs to overcome in order to be successful. I will list some of the challenges in this article.

1. Architecture -

There are two types of SoCs generally in today's smartphone market. Most SoCs such as Samsung's Exynos and Mediatek simply use the stock Cortex cores provided by ARM. Although ARM's Cortex cores are good, its entirely possible to license the base architecture and build custom cores that are even better. This is exactly what Qualcomm and Apple do. Qualcomm licensed ARM's v7 architecture and created Krait custom core. Apple licensed ARM's v8 architecture and has created Cyclone, Typhoon and Twister custom cores.

The custom cores developed by Qualcomm and Apple were used in their SoCs. Qualcomm used Krait in various models from Snapdragon S4 to Snapdragon 805 while Apple used Cyclone, Typhoon and Twister in A7, A8 and A9 respectively.

Developing these custom cores requires a lot of effort and resources. It often costs companies hundereds of millions of dollars but they are necessary if the company wants to make its SoC stand out in the crowd. Developing these custom cores enables companies to drive better performace at lower power consumption. This has certainly been true for Apple whose A7,A8 and A9 have been at the top of benchmarks recently.

Should Google try to enter the SoC market it definitely needs to develop a custom core of its own to make its SoC stand out in the crowd. But as said earlier these custom cores require a lot of effort. Apple's acquisition of PA Semi happened in 2008 but it wasn't until 2013 that Apple actually got the A7 SoC out which truly showcased its prowess as a SoC designer. Similarly when Apple launched A7 SoC in 2013, Qualcomm was caught flat footed with them having no custom core based on ARM's v8 architecture to rival Apple's A7. Its only towards the near end of 2015 that Qualcomm has finally released Kyro which is its custom core based on ARMv8. Between 2013 and 2015 Qualcomm had to rely on stock ARM cores which jeopardized their performace severly especially in case of SoCs like Qualcomm Snapdargon 810. This lead to Qualcomm losing Samsung as its cutomer which lead to a great finacial disaster for Qualcomm.

So its pretty clear from all this that designing a custom core requires.

A. A lot of money
B. A lot of talent
C. A lot of time

Only with the proper combination of the three mentioned above can a company get a competitive SoC out in the market.

2. Competitive edge -

Although having a custom core is of great importance, its also necessary for the SoC maker to have a competitive edge when compared to others in the market. Sure the competitive edge almost always means delivering the maximum amount of performance for every watt of power consumed but apart from CPU, an SoC consists of lot more such as GPU, Modem etc which also have great importance.

Qualcomm's competitive edge for example has been its ability to integrate the modem along with the SoC. This has allowed Qualcomm to have a monopoly of sorts in the North American market. In North America, smartphones are sold mostly via carrier stores and undoubtedly then do carriers even get the option to decide which smartphones get stocked in the carrier stores and which don't. Often the No.1 criteria carriers look at while certifying devices is the SoC at use and Qualcomm's the only ARM SoC manufacturer that has global carrier certification. The global carrier certification combined with Qualcomm's patents on CDMA make their modems one of the best. This has also led to Apple using Qualcomm's modem in the Ax series chipsets recently.

Apple's speciality with their SoC is deep integration. The best example of this is Touch ID where Apple created something called the secure enclave in the A7 SoC that stored the fingerprint data .

Similalry Nvidia's speciality is in GPU, their SoCs have often been used in tablets and are considered one of the best for gaming.

Mediatek and several chinese SoCs are great at cost cutting, they provide some amount of decent performance at prices cheaper than what Qualcomm would ask. These feed the bottom end of the smartphone market.

Were Google to make their own SoC they even need a speciality feature to make their SoCs even more competitive.

3. Target market -

Samsung makes its own Exynos SoCs because the flagship Galaxy S and Galaxy Note device often sell in millions every year giving Samsung enough incentive to develop their own SoC in case Qualcomm fails to impress as seen with Qualcomm Snapdragon 810.

Qualcomm given its expertise in modem and the treasure trove of CDMA patents the company holds is by far the only option for companies like LG, HTC, BlackBerry, Microsoft and Sony that want to sell their devices in North America. Also given their expertise in LTE Qualcomm has a quite a few mid range and low end SoCs such as QSD410/615 etc to cater the demands of emerging countries such as India and China where 4G networks are emerging rapidly.

Mediatek and other Chinese SoC makers keep serving several Chinese smartphone manufacturers and smartphone manufacturers of developing countries. These smartphone manufacturers target the lower end of the market and the cheap prices of Mediatek, Spreadtrum etc help them.

Apple of course ships millions of iPhones and iPads which are a natural fit for its Ax series SoCs. Apple's Ax series of chipsets also helps Apple take integration to a whole new level as seen with stuffs like Secure Enclave which was applied with Touch ID or Apple's custom DSP which helps them do some great post processing of photos and vidoes.

The reason I am mentioning all this is because almost all major SoC makers have markets established for them. Also the smartphone market isn't having the kind of growth it once possesed especially so in developed markets such as US, Japan, UK etc, hence Google will have to snatch market share from others if they are to be a major SoC maker market share wise.

4. Return on investment -

This is a very important topic. As mentioned earlier designing an SoC requires lots of money and companies need huge scale to be able to recover their costs. Qualcomm, Mediatek and Apple are often able to sell hundreds of millions of SoCs every year which helps them reap back their investments. Margin also plays an important role here, the devices using Apple's Ax series of chipsets carry heavy margins as compared to something that has a Mediatek processor.

For a SoC makers to operate profitably, scale is very important. Google lacks this scale. As mentioned earlier Google would have to get market share from scratch. Currently the only Android hardware that Google sells are Nexus and Android One and both of them have very low scale and definitely nowhere enough to help Google reap their investments in case they decide to create an SoC on their own. It also remains to be seen as to how well Google's able to attract other manufacturers to use a Google made SoC, but if I were to take a blind guess I would say very few would. Manufacturers are already seeing a race to bottom situation with a majority of them using Android and having very little differentation, they wouldn't want differentation to become even lesser by using Google made SoCs I presume.

5. Foundry -

SoCs are designed by companies like Apple and Qualcomm but Apple and Qualcomm don't own any foundries as such. Foundries are in a way mutli-billion factories where SoCs are churned out. Space in these foundries is limited and currently only Samsung and TSMC are the major foundry players in the industry. These foundries bring out new process nodes on which SoCs are manufactured. A newer process node means better performance at lower power consumption but capacity for every new process node is limited.

Most SoC designers want the latest process node for their designs so as to bring out the best in their chips but because of limited capacity only few designers are able to get the latest process node. Since Samsung itself owns foundries and Exynos is also made by Samsung, Exynos often gets manufactured on latest process nodes. This was very evident with Exynos 7420 that showed considerably better performance than Qualcomm snapdragon 810 despite architecturally being the same. This was majorly because Samsung's Exynos 7420 was made on 14nm process whereas QSD810 on a 20nm process.

Should Google make their own chips even they would have to compete with Apple, Samsung, Qualcomm and tons of others for foundry space.