Google hires a key Apple chip designer | Hacker News Reader

Google hires a key Apple chip designer

engadget.com

122 pointsjrwan73 comments

"Right now, its only in-house silicon is the Pixel Visual Core imaging chip inside the Pixel 2"

This statement completely ignores the group making the TPU. Google has more hardware expertise than just the consumer electronics folks.

There's a lot more. The article makes it sound like this is all recent, but the Visual Core was in the works for at least three years. Then there's the Titan security chip and the Lanai processor used in Google NICs. Google is heavily involved in RISC-V, through David Patterson, who also worked on the TPU and is, of course, famous for the Berkeley RISC that eventually turned into Sun's Sparc. It also has expertise in other areas like memory chips or more esoteric ones like silicon photonics. Then there is a bunch of former PA Semi folks and even an Alpha architect, Dick Sites, although none of those worked on CPU design at Google, AFAIK.

Patterson like Hennessy & Patterson?

Yes, same David Patterson.

Yes. Hennessy is a Google board member, too. It's a small world.

Dick retired last year. But your point still stands, as Google is flush with academic and industry veterans in processor design.

Also Norm Jouppi

I thought LANai was a Myricom design, not a Google one.

Originally, yes:

https://www.theregister.co.uk/2016/02/09/google_processor/

There's no real sources there that say that Google has custom LANai silicon.

And, from what I've seen when the register writes about subjects that I actually have signed the NDA for, they just blatantly make shit up, knowing that no one is going to break their NDA just to correct them.

AFAICT, Google was just given firmware access to Myricom's chips, and wrote an LLVM backend.

There are quite a few ex-Myricom people at Google, including two of the founders. One of which was also CEO: Nan Boden's bio mentions joining Google via acquisition.

https://www.crunchbase.com/person/nan-boden

So? Google didn't acquire Myricom, CSPi did.

The exodus to Google (including the CEO and CTO), which Nan Boden's bio itself describes as an acquisition, happened months before the sale to CSPi. Draw your own conclusions.

Boden heads in a division getting businesses to use Google services, not internal infrastructure or hardware.

Plus Google hired Norm Joupolli (sp?) who worked on the MIPs chip a couple of years ago.

To be honest though this is exactly how Google likes it. Google actively encourage external production of misleading data about their capabilities.

I recall sharing an article around inside of Google where the journalist had mentioned the development of a new data center and speculated that Google had at least 3 and as many as 6 data centers that they had built. At the time they were spending a billion dollars a quarter ($4B/year) on "capital expenditures" :-)

Would tend to say the opposite. Google kept the PVC a secret until found in the pixel 2. We did not know about the TPU until used in production for a while. We did not know about the gen 2 TPUs until done. There is a long list of Google doing things and never sharing. Kind of the opposite of what you suggest.

I agree with you, I think I said it badly. Google is very secret about what they do and what capabilities they possess. If Urs had his way nothing they ever did would ever leave the building :-). But there are times when circumstance leads to a hint of what they are doing to leak out.

When that happens, Google is pretty good at getting the narrative out that misleads about their actual capability. The real capability can be 100x what external sources might lead you to believe.

Can anyone layout a path to becoming as skilled in chip design as these people like John Bruno? As though I would be able to do a competent Verilog implementation of RISC-V or 68040 in my own workshop.

As a hobby (back when I had the luxury of time) I designed several ALUs, and two CPUs featuring datapaths and register transfer control. I did them as digital circuit diagrams on paper, on breadboards (which teaches the valuable lessons of signal fanout, wire layout, debugging), and in squirrel (which teaches the value of good tooling, one way or another). I've toyed with VHDL, Verilog, and FPGAs, and am competent in assembly and C. I'm interested, but I don't know what the path forward looks like to gaining the chops that these famed designers have.

> As a hobby (back when I had the luxury of time) I designed several ALUs, and two CPUs featuring datapaths and register transfer control. I did them as digital circuit diagrams on paper, on breadboards (which teaches the valuable lessons of signal fanout, wire layout, debugging), and in squirrel (which teaches the value of good tooling, one way or another). I've toyed with VHDL, Verilog, and FPGAs

Not an answer to your question, but I'd love to read some blog posts about this please.

That's the luxury of time thing.

If I was a recipient of universal basic income, it's the first item on my list. A series starting from grass roots leading to code-level manifestation and verification of landmark chips like MOS 6502, Zilog Z80 and onwards up to MIPS R4000, SuperSPARC, Intel 80486, Motorola 68040. And a deep dive to implementing a lisp-based OS and tooling to both run & compile itself, completing the circle. There is so much interesting about CPUs, from large architectural design considerations like machine type (Von-Neuman or Harvard) and MMUs through to minutiae like choice of instructions and the encoding of the instruction set, balancing the conflicting demands of instruction timing optimizations versus silicon space and power budget, math intrinsics, ... this rabbit hole runs deep. I'm starting to wonder if any one person is able to master all of the requisite topics, similar to the pseudo-fact that there is no one person who is able to manufacture a can of Coke from scratch. But, we can try, and that is where the fun is :)

Google and Microsoft both expressed concern about the potential Broadcom takeover of Qualcomm. And since Broadcom and Apple on on good graces one can't help but wonder Apple's influence in all of this especially when you consider the timing. If Broadcom does take over Qualcomm then Google needs to be prepared to pivot in the event Broadcom decides to take actions that are not aligned with Google's interests. I'd also like to see some real innovation in Android SoC's because there certainly hasn't been any from Qualcomm who's R&D strategy is now taking ARM's latest reference design, making a few modifications and calling it a day.

Google had earlier (well, 7 years earlier) bought AgniLux, a follow on company with people from PA Semi which had been bought by Apple. That looked an awful lot like an acquihire. Dunno if they went on to do TPU or what.

https://arstechnica.com/information-technology/2010/04/googl...

They worked on Chrome hardware. Google is heavily involved in the development of the boards that end up in manufacturers' Chromebooks and Chromeboxes.

The cycle is complete: DEC and IBM are reborn, and the Intel business model is in retreat.

AMD's tragedy, the make brilliant engineers in house and have talent at picking good cadres, but very few of them will go to work on any greenfield development for them.

The moment a cadre on a level of Keller or Bruno gets bored, they loose them

I wonder if they plan to build an ARM CPU or a RISC-V CPU.

https://riscv.org/membership/999/google/

No, they are working on their own SoC to launch it in the next Pixel phone or in the one after.

They want to become like Apple and control their entire ecosystem all the way from the hardware to the software in their premium phones. They are tired of the current mess with all smartphone makers polluting the market with Android phones that stop being supported after a couple of years. Their target to have a full premium Google phone is around 2019.

Source: wink wink

Seems a bit slow on the uptake. It was clear in 2013 from what Apple did with the 5S that in-house CPU was a big win.

Do they also plan to cater to the lower end.

The Apple A series processors are ARM based...

I think one of the confusing things is ARM holdings doesn't sell physical chips. They sell a spec (verlog code I believe) that can be turned into a chip. Add some usb, memory, video card, maybe wireless and you get a system on a chip.

In short - ARM licenses its architecture for which you pay a fee.

Except that it isn't short. ARM licenses their architecture which you can implement but that implementation also must pass a verification test and you can neither add anything nor leave anything out. ARM also licenses IP cores in its traditional fabless role.

There are about 15 architectural licensees. Apple and NVidia are architectural licensees.

https://semiaccurate.com/2013/08/07/a-long-look-at-how-arm-l...

Samsung, Qualcomm and Texas Instruments license cores.

https://en.wikipedia.org/wiki/Arm_Holdings#Licensees

ARM sells many things, that many different customers use.

One is a license run the ARM ISA, so you can design a chip from the ground up. Another is a license to produce a reference implementation of a single processor design. Others peripherals and options exist too.

So, in short, a microprocessor architect might be needed to build an ARM chip, depending on how one was going about it.

and to be clear, what Apple has is the ISA license. Apple's ARM CPUs are not just the reference design from ARM. I'm not enough of a hardware person to know if they are modifying the design or have implemented the ISA from scratch, but their chips seem to be significantly faster than anyone else's so they're clearly doing something smart :)

They've almost certainly designed their cores from scratch.

The funny thing is that when Acorn spun off ARM some 25 years ago, Apple owned a chunk of the new company, since they used ARM6 for the Newton. The new chip's MMU was actually designed by Bob Welland, one of the fathers of Newton OS, for which he concocted the whole protection domain feature.

It seem fairly unlikely that they'd try to come up with their own CPU architecture for this.

My Pixel XL is the first Android phone I could see lasting me long enough to get to 2019.

Same here with my pixel 2 XL .. I hope to keep it at least until 2020, maybe a bit more (with a battery replacement in the middle).

I would have said the same of my Nexus 6P .. but it has many hardware problems that manifest after a while. Fingers crossed to do not see the Pixel do the same.

Same with my Pixel 2 XL. So far the best smartphone I have owned. I now prefer Android over iOS and the Pixel 2 is as smooth as any Android phone I have ever used.

I hope you are right because I feel that the Android ecosystem is a quagmire.

Unlike my Google Moto G that was unsupported after 18 months?

The problem with the current phones is that they are dependent on the SoC's designer for update support. If Qualcomm decides to phase out support for one of their SoCs, there is nothing that Google can do about it.

This is the kind of situation that they are annoyed with.

I though this was no longer a problem with Project Treble?

Project Treble will improve the situation indeed but Google is still keen in having control of the ecosystem like Apple.

It gives the tool for new oreo phones to handle this correctly, but it is still up to the OEM to do it.

It’s not just the carriers and manufacturers, Google doesn’t even support their own phones beyond a few years (e.g. the Nexus 5).

3 years of OS and security updates is actually quite good for an Android phone relatively speaking. Qualcomm doesn't usually agree to support their SoC's for more than 2 years so getting an extra year out of Qualcomm must have been expensive. And therein lies the problem - Google has been beholden to Qualcomm to update their drivers for every major release of the OS. Perhaps this will change now that the Linux kernel has 6 years of LTS and Android 8 and onward now allow the OS to updated without modifying the kernel, but who knows.

One interesting note is the length of support Intel has provided to the Nexus Player. First released in 2014 on Android 5 it recently received Android 8 in 2017. It's too bad Intel is out of the mobile SoC market because their level of support has been unparalleled by any SoC manufacturer.

It has been a rumor for ages but so far we are still waiting for this SoC :/

I'd be worried about them doing to RISC-V what they did to Android: buy it, make it popular, add proprietary opcodes that substantially improve performance, get all major RISC-V chips in production using those opcodes, slowly change the spec and ecosystem over time to make the original open source project completely obsolete.

If you have a Pixel 2 XL, you can:

Download the source code for its Linux kernel: https://android.googlesource.com/kernel/msm.git/+/android-4....

Compile it:

  export ARCH=arm64
  export CROSS_COMPILE=aarch64-linux-gnu-
  make wahoo_defconfig
  make -j$(nproc)

Build the userspace by looking up the correct source code : https://source.android.com/setup/build-numbers (e.g. branch android-8.1.0_r2):

  mkdir $AOSP_HOME
  cd $AOSP_HOME
  repo init -u https ://android.googlesource.com/platform/manifest -b android-8.1.0_r2
  repo sync --jobs=$(nproc)
  source build/envsetup.sh
  lunch aosp_wahoo-userdebug
  make -j$(nproc)

And then flash your phone by doing fastboot -w flashall, you'll have root access and everything. (omitted downloading some kernel modules for the hardware)

This post is pure whitewash bullshit.

"omitted downloading some kernel modules for the hardware"

aka all the drivers, the actual things that make the hardware work, which are all closed source, and eventually will stop being updated (likely about 6 months from now) leaving you on a forever obsolete android release and kernel, full of exploitable holes.

Project Treble makes it so drivers do not get obsoleted

Not at all, Project Treble makes it so, assuming that OEMs upstream their drivers, but as usually happens in the whole fragmentation story Google is only advising OEMs to do it, not enforcing it.

People need to realize Project Treble will only improve Android, if Google has the guts to force OEMs to play ball, until then having a certified Treble device doesn't mean updates will be made available.

You can't say it won't happen until it actually doesn't happen.

And it appears to be happening: https://www.xda-developers.com/project-treble-aosp-android-o...

It only counts when random dude on the street is able to do it via then update OS menu option.

Being possible by people with technical skills to replace firmware doesn't count on my book.

I care about the regular dudes buying regular pre-pay phones that they can afford.

I was responding to the comment by tfha. The situation could be far worse.

Aside from the obvious clickbait title, the article is very light on detail and filled with speculation. The original referenced article is here (paywall):

https://www.theinformation.com/google-ramps-up-mobile-chipma...

Doesn't the word "poaching" in the title imply that the employee was somehow Apple's property? Wouldn't "Google hires a key Apple chip designer" be a clearer, less sensational way to put it?

I mean, the employee had leased/rented/sold/whatever their labor output to apple and now google has taken over the lease/deed/contract/whatever.

In a way, your employer does own something: rights to the output of your labor.

Strictly speaking, yes, but it's a well known and used term to refer to the practice of somehow hiring an employee away from a company.

The term is demeaning to employees, but the connotation is pretty well understood. It's not really meant to convey proper ownership, just that the employee was lured away.

the article doesn't make it clear but they're actually cooking him in boiling water.

On a safari

More like chrome now

It used to mean that, but the term's lost its meaning a long time ago, unfortunately.

People can't even bring themselves to use the distinction for Waymo/Uber and Levandowski.

Not sure what you mean about Waymo/Uber. Are you suggesting it’s just an act of poaching as opposed to theft of tons of coincidental intellectual property and industrial espionage?

It's what the headlines would read if this was about a sports star.

>> It's what the headlines would read if this was about a sports star.

This is probably the correct interpretation. "Poaching" seems to be the word used when a person is specifically targeted because of their experience. They're viewed as key, so it's not just good for the one hiring them, but bad for the company losing them. If you think of hiring as dragging a net through the water an sifting through the catch, poaching would be more like spear fishing.

Or it could just be sensational headline writing.

is Google and Apple only companies in this whole world anymore can create mobile phones ? Why is whole world following these 2 companies ? Is this the "internet / mobile phone" ghetto ?

Almost like google and facebook, is there anybody who could leave these services and understand, they are created to spy you, just like your mobile phones.

well, purism is trying, MSFT tried, Jolla tried.

mobile is a tough market to crack! its expensive and the network effects that influence what people buy are pretty strong.

Google and Apple have the only two mobile platforms so of course that is where focus will be.

Reply on news.ycombinator.com