Pixel 9’s Tensor G4 will not be the bold chip Google wished

When Google launched its first Tensor chipset with the Pixel 6, its intentions have been fairly clear: to develop unbiased of Qualcomm’s Snapdragon (or some other chipset vendor) and to construct experiences that might in any other case be not possible. Whereas we’re nonetheless ready for the Pixel 8 sequence and its accompanying third-generation Tensor G3 processor, Google’s chips have already delivered some spectacular options that might probably be not possible on different chipsets, particularly in AI and picture processing.

All of this, nonetheless, was considerably undermined by the truth that Tensor chips have, to this point, been designed in collaboration with Samsung. Whereas they function components constructed by Google, main elements of their design are derived from Samsung’s Exynos chips.

Sadly, this may stay the case for the upcoming Tensor G3 and future Tensor G4. Because of a Google insider, Android Authority has now realized Google’s Tensor plans for 2024 and past.

As reported by The Data (paywalled), Google initially deliberate to launch its 2024 Pixel sequence with a “absolutely {custom}” Tensor SoC codename “Redondo” (internally also called “RDO”), constructed on a TSMC course of node. Nonetheless, as a result of a missed deadline, the chip was too late to be included in 2024’s Pixel 9 sequence (as chip improvement is a prolonged course of, often beginning years earlier than launch).

Google has as a substitute determined to fabricate the SoC for testing functions, as its successor, which can probably make it into the 2025 Pixel units, will share lots of its design parts. As we’ve realized, Google has designed a improvement board known as “ChallengerDeep” to work on the chip.

There may be, nonetheless, nonetheless the matter of Pixel 9. With Redondo out of the image, Google wants one other chip to grow to be the Tensor G4. The answer to this downside is a brand new chip codenamed “Zuma Professional.” Observe that the Tensor G3 — the SoC that’s coming to the Pixel 8 sequence — is named “Zuma.”

The brand new chip remains to be semi-custom, co-designed with Samsung’s System LSI division, and can probably be a smaller improve than initially deliberate. That is just like how Tensor G2 (codename “Whitechapel Professional”) was a modest replace over the unique Tensor (“Whitechapel”). The chip at present runs below a improvement board codename “Ripcurrent 24” (additionally known as “Ripcurrent Professional”), whereas the Tensor G3 used simply “Ripcurrent”.

Google can be already engaged on the next-generation fully-custom Tensor chip codename “Laguna Seaside” (or just “Laguna”) for 2025 Pixels. The event board for the chip is codenamed “Deepspace.”

You could be confused about what “absolutely {custom}” or “semi-custom” even means within the context of SoC design. To reply that, we have to have a look at the historical past of Tensor chips and the way they’ll change after this swap.

Co-designing an SoC with Samsung

Google most likely first approached designing its personal SoC one thing like this: a number of years again, it wished one thing to combine the nice work of its current {hardware} groups (just like the Pixel Visible Core, the Titan safety chip, and the Edge TPU) right into a single smartphone chip that might be uniquely tailor-made to its wants and necessities.

At the moment, the Google chip staff (referred to as gChips) was pretty small, so designing a whole SoC from scratch could be troublesome (if not not possible). As an alternative, Google determined to search out somebody who might deal with components of the design course of that it didn’t have the capability for whereas its staff labored on elements that extra instantly affected the Pixel consumer expertise.

Samsung LSI reportedly established a brand new “Customized SoC” division in early 2020, offering shut design assist to new chipset clients. The corporate was trying to leverage its large library of IP blocks, encompassing CPUs to 5G modems, counting on its confirmed monitor file in a large number of merchandise. We don’t know precisely when Google approached Samsung (it might probably have been in 2019 or earlier), however the nature and timing of Tensor and public consciousness of Samsung’s {custom} chip enterprise line up fairly nicely.

Google may have delivered an inventory of its challenge necessities, consisting of the {custom} {hardware} it wished to incorporate and particular customizations — for instance, Tensor’s extra distinctive 2+2+4 CPU core structure. Samsung started engaged on its a part of the contract, utilizing IP from its huge library to construct out the remainder of the SoC to the precise necessities. Google, in the meantime, began making ready to ship its {custom} {hardware} components (such because the Titan chip, Edge TPU, {custom} ISP parts, or a {custom} AV1 decoder) to Samsung for the ultimate integration. After that, Samsung started engaged on the bodily design of the chip and ultimately manufactured it in one in every of its fabs.

After that, Samsung helped Google within the first levels of labor on the chip (known as “the bring-up”) and additional assisted them with fixing points and bugs. The corporate additionally supplied Google with software program for a number of the design parts which might be within the chip.

Because of all this, Google was capable of get its first {custom} chip available on the market in significantly much less time and with a smaller staff than it might have taken to construct Tensor independently.

Over the previous few years, gChips has drastically expanded its staff and expertise. It has additionally expanded its IP block library. For instance, Tensor G2 launched a DSP known as GXP. Google is more and more able to ditching Samsung’s hand-holding.

So, what precisely modifications with Redondo and Laguna Seaside? Google will now be answerable for all of the components of the design course of beforehand accomplished by Samsung.

It’ll first should supply the required {hardware} IP blocks, be it by designing them in-house or by licensing them from a third occasion. There are plenty of firms offering a lot of these providers, like Arm, Cadence, Synopsys, or… Samsung. It’s fairly probably that Google will nonetheless use some IP initially utilized in Exynos chips in its “absolutely {custom}” chips, however the nature of the design will probably be much more customizable.

Google’s design staff may also should cope with the verification and bodily design of the chips, two notoriously troublesome elements. Not solely that, but it surely’ll be answerable for coping with chip fabs (like TSMC) and different miscellaneous steps they may encounter within the course of. Loads of software program will probably must be written from scratch, and Google will now not have Samsung to ask for assist.

How is that going to look in apply? It’s unlikely we’ll see drastic modifications to the upcoming Tensor chips within the brief time period. Transferring to totally {custom} is a bid for independence and can enable Google to customise its chips past what’s at present obtainable, but it surely’s not a miracle — it’s merely a logical subsequent step within the improvement of Tensor. One factor that we all know for positive, nonetheless, is that it’s going to allow them to swap to TSMC’s foundry, which at present presents manufacturing nodes much more environment friendly than something supplied by Samsung.

We’re excited to see what Google has in retailer for its future Tensor chips. Whereas Redondo being delayed is undoubtedly disappointing, it’ll in the end result in an excellent higher Tensor G5, which needs to be nice for future Pixel patrons.