AMD Zen microarchitecture and Intel's Ocean Cove patent | Hacker News Reader

AMD Zen microarchitecture and Intel's Ocean Cove patent

twitter.com

365 pointsParseus90 comments

Remember the advice of "How to read a patent in 60 seconds" [0]. In this case there are two very short independent claims (1 and 3), all the way at the last page. They both seem to just be talking of some specifics on cache coherency.

[0] https://www.danshapiro.com/blog/2010/09/how-to-read-a-patent...

While this might be true from a legal standpoint, even using a competitor's presentation for the description seems a bit lazy, and getting caught at it is quite embarrassing. It implies that nobody at Intel can describe a modern processor very well.

It's definitely lazy that they didn't create their own description, but how does it follow from this that they can't?

If it's published, it is prior art.

They are not trying to patent what's shown in the figures, those are just given as background information. A patent for a screwdriver will most certainly also describe screws, regardless of whether screws are prior art.

Patentability is solely a matter of the claims made at the end. Those claims will of course be interpreted in the context set out by the method description, but it is entirely normal for a patent to describe the state of the art before discussing a novel method (about which actual claims are then made).

(IANAL)

It's still blatant copyright infringement at least.

IANAL but I would have thought that referencing material for prior art would have fallen under fair use.

Yeah, to me it just implies they don't want to actually leak any confidential information about their architecture since those details aren't really relevant to the patent (which is about snooping cache line zeroing in a cleaner way), so they just described their competitors overall uarch. Stealing lines from an anandtech article is a line to far though IMO.

Patents are generally not written by engineers. They are written by patent attorneys / outside consultants. The engineers have a few conversations with the attorneys, and share technical material (in this case, something specific about cache coherence). The attorneys/consultants then generate a hundred pages of patent-ese scaffolding.

The plagiarized content:

1. Is not what is being patented here.

2. Is not created by Intel engineers (or likely even Intel employees). They're busy designing chips.

Ian Cutress (discussed elsewhere in these comments for also having his description of a CPU copied in the patent) suspects it's an intern[0]

[0] https://twitter.com/IanCutress/status/1511716678199132173

With proper citation, what's the issue with an article being quoted?

It's a common tactic in patents in my experience. You're not going to sue yourself. So you describe your invention in the context in which a competitor would use it. Of course, this would typically be coupled with a disclosure to the examiner that the thing describing the context is in the prior art.

> It implies that nobody at Intel can describe a modern processor very well.

This seems like a bit of a stretch, given that Intel had a lead in CPU power for years and is still quite close to AMD.

Designing a modern CPU architecture and describing one to a non-expert audience are very different skills.

Intel is citing AMD as prior art. It isn't lazy. It's necessary background.

Engineers don't write the text of patents, patent attorneys do. And for the kind of boilerplate in a patent like "describe a modern processor", this text and images are actually quite likely to itself be heavily copy-pasted from previous applications.

Honestly, I would be more surprised if any engineer (intern or otherwise) at Intel ever noticed that the images came from an AMD presentation, let alone actively participated in copying the images from said source.

As the inventor on twenty-something patents, I think you're understating how much input engineers have. Yes, the bulk of the work is done by patent attorneys, mostly translating from pure technical jargon into that weird mixture of jargon and legalese that patents are written in ("...of the one or several..." and "...such as but not limited to..." and all that rot). However, for a patent of any complexity this often involves multiple rounds of back-and-forth with engineers to clarify and ensure correctness. In fact, I'd argue that any engineer who isn't involved in the writing process is failing to be sufficiently diligent or act in good faith when they sign the part about the patent accurately reflecting the invention.

Nuance: the solution to all things ever.

I probably wouldn't have spotted this.

It looks suspicious because they reproduced the diagrams using what looks like a fucking typewriter.

USPTO will actually search for this kind of thing (assuming the situation is as-described, which I suspect is not assured). Source: this happened to me -- I wrote an internal doc describing a new product feature. That doc was used both our product documentation team has the basis for user documentation; and by our patent team as the basis for a patent application. When the patent was reviewed, the examiner found the user doc on the internet and rejected the application due to prior art. We had to prove to the USPTO that both were derived from an older doc (by finding the older doc with timestamp and sending it to them with associated legal knobs and whistles).

At first glance this doesn't look like an exaggeration. It's basically a 1-to-1 rip off when you place the images side by side. Even the labels within the diagram are copied directly.

Those slides aren't the only thing they copied: https://twitter.com/IanCutress/status/1511718827083669508

So they cited Ian's 2016 article (which contained the AMD diagrams), then basically rendered those same diagrams in black and white in the patent. Pretty lazy.

You never use StackOverflow or similar?

A lot of people use StackOverflow but people blindly copying StackOverflow is just as dumb. When I use it, I use it to understand the problem. I then write my own logic understanding the issues wrapped around the similar error. But some solutions are too simple to not be similar.

Obviously most devs will answer yes to this question. However, I don't think the implied conclusion (software devs copy each other, so the legal apparatus applied to software is okay doing so, too) is a valid one.

For one, this is hardware.

And second, even if this was software, that just reinforces the farce that is software patents -- another reason to add to the list of why patents on software shouldn't be a thing.

That's how the US patent system works now.

"The first inventor to file (FITF) provision of the America Invents Act transitions the U.S. to a first-inventor-to-file system from a first-to-invent system and became effective on March 16, 2013"

https://www.uspto.gov/patents/first-inventor-file-fitf-resou...

To be clear, first-to-file just means that in the event that the patent is granted, it's awarded to the person who filed first instead of the "actual" inventor. However, the patent can still be invalidated for prior art. The net effect of this is that AMD lost their ability to patent their invention by not filing first.

Perhaps this was the entire point of Intel's work here??? Worst case scenario for Intel they block AMD from patenting (and allowing Intel to use it).

Presumably if AMD hasn't patented it by now, they weren't ever going to patent it. CPUs take years to develop, so by the time something shows up in marketing slides (ie. release is imminent in a few months) it would have been years since the idea was first conceived.

And first to file doesn't mean first to file with the USPTO.

If you have a great idea for air-conditioned socks, you file your initial notes/work with your patent attorney, and that is sufficient (general hand-waving about officer of the court and so on.)

> The net effect of this is that AMD lost their ability to patent their invention by not filing first.

It can explain Intel's move to file the patent, which seems stupid at a first glance.

Anti-Hanlon Razor: never attribute to a stupidity anything that was done by an intelligent entity.

Perhaps known as Gray's Law:

“Any sufficiently advanced incompetence is indistinguishable from malice.”

In particular this interpretation [1] may be best applicable

"In other words: If you find someone acting so stupid that you can’t believe she’s doing it without the slightest chance of knowing that it’s stupid, then she might act out of malice."

[1] https://agiletrail.com/2011/12/20/hanlons-razor-comfort-in-t...

something I’ve wondered for a while:

Suppose someone invents something and uses it for their products, but never publishes anything about the invention, and then someone else (years later) independently develops the same thing and gets a patent that covers the first invention.

Is the first person now infringing the patent, or can the fact they were doing it first (even though there is no published info to serve as prior art) give some rights to keep using it?

(I understand any responses are not legal advice and I should ask a lawyer, etc, etc)

‘It depends’ - but generally the first person is now infringing.

IANAL. This is my current understanding. If I'm wrong, feel free to correct.

Using it in a product counts as "public disclosure". If you've shipped it, that's a public disclosure for patent purposes.

But it's more complicated than that, because you can file for a patent on something within one year after the first public disclosure. That also means that you can file within one year after someone else publicly discloses something, if you lack morals and decency.

So if person A develops something and starts shipping it, and person B files more than one year after the first shipment, then the device is now "prior art" that invalidates the patent. Person A faces a legal fight, but they shouldn't be infringing, because the patent should be invalid.

Just to be clear, in the US you must be an inventor to file. You can't take someone else's invention, once publicly disclosed, and then patent that within the one year period. But if two people have same invention the first to file usually has precedence.

Of course in practice there are thousands and thousands of trash patents but technically those are the rules.

Relevant statute:

https://www.law.cornell.edu/uscode/text/35/273

Real answer though is "it depends", so ask a lawyer about specific facts.

If first person has proof through dated notebooks, schematics, or internal documents they would likely be fine. However, second person may still have valid patent to assert elsewhere if there was never anything qualifying as public disclosure by first entity. Particularly under current doctrine.

You seem to think AMD has no patent on Ryzen. Don't worry, they have. Hopefully not containing loads of diagrams from Intel :D

That isn't what first to file means. First to file doesn't change AMD's publication from being prior art. You cannot create a valid patent by filing patents on other people's publications.

It seems to me the Intel patent is using the AMD designs as examples, not something that's actually patented.

The tweets only show illustrations, not the text of the patent. So that's pretty plausible.

It is not pretty plausible. It is fact. The claims are strictly about cache. It barely even mentions the architecture except when relevant.

...I think the actual patent is on moving the cache snooping into the interconnect?

I think all of the figures presented are part of Intel's claim that any processor that implements cache snooping in the interconnect are subject to this patent.

Sorta. Cache snooping has been part of the interconnect for years (it arguably is the interconnect), but this is about snooping cache line zeroing instructions without having to go out to memory as far as I understand it. So if a line is shared in two caches, it can be zeroed in both without getting the DRAM controller or even L3 involved.

These diagrams could practically describe any processor, they are not even x86 specific. The fact that the layout of the figures is similar is kind of suspicious but I would guess they just come from the same decades-old book (or patent). They do reuse these type of quasi-standard figures quite often in order to churn out patents faster and faster. You'll find the actual new content (probably minuscule) somewhere in the claims. The patent system is just that ridiculous.

The specific numbers also match, but I agree that the diagrams are fairly generic.

Definitely a lawsuit is going to come of this. I'm not even talking the blatent copy of the presentation but the blatent patenting of memory on interposer. This is basically what AMD's 2016 patent encompases for much of Ryzen's design architecture especially with HBM.

Now I think where Intel might be in the right is that they will claim improvement on AMD's patent and design. But it will be a stretch considering they are using AMD's presentation. Which obviously in court it will incredibly hard for Intel to prove no wrong doing.

NAL but you have to check your patents before submitting them. Intel went full patent troll.

The “NAL” statement wasn’t necessary: it was totally obvious the moment you wrote “definitely.”

There’s a 99.999% certainty that this non-issue will NOT result in a lawsuit.

Gonna NAL anytime it comes to such subjects. ;)

Amazon has a full team of patent lawyers on retainer just to submit a patent.

> Amazon has a full team of patent lawyers on retainer just to submit a patent.

So does any larger tech company, no?

It turns out that this "story" was intentional misdirection, posted to test how tech information is misinterpreted and spread. Intel has not patented something of AMD's design.

Underfox writes:

>Now that a sufficient amount of time has passed and I have collected a sufficient amount of data, let's explain today's social experiment.

>For those who don't know, in addition to my work with telecommunications, I'm also involved in researching complex networks, where among many lines of research is the way information propagates in social networks.

>With the data obtained today, I was able to help a great friend with his research and, at the same time, show how technology information is misinterpreted and why filtering is so necessary.

https://twitter.com/Underfox3/status/1511865763132542982

Haven't AMD and Intel had to share all their patents with each other for the past 20 years anyway? I remember some agreement coming out of the ashes of the amd64 debacle. And the patent text doesn't seem to have anything to do with the images.

Only for x86 up until that point in time.

Unfortunately the author chose to show his arguments in the Twitter. The subject is interesting, but I have a hard time following the discussion there.

So... Would the patent claim be a copyright infringement of AMDs presentation?

AMD could invalidate it as prior art pretty easily.

How so? The claims have very little to do with the diagrams. They are only there as an example processor for the claims being presented. The claims are specifically about cache. More specifically clearing/invalidation. What better way to prevent the competition from implementing something similar than using the competitions architecture as the base to show how it could be implemented? It is Intel's prior art against AMD.

As prior art. You cannot patent things which are not your invention and a prior knowledge. Would be hard to proof it is your invention if you use knowledge from your main competitor in the patent application

They are not trying to patent anything other than their own work. Their work is something which can be added to certain processors(AMD's included). Why not base the example on a competitor's processor so there is less of a chance someone could argue it is reliant on Intel's processor architecture?

Why would one ever do this as a requirement for patentability is that it doesn't exist as prior art?

Because they're not patenting the stuff in the figures.

This is the equivalent of plagiarizing the "related literature" section of a paper - lazy, in bad taste, but need not invalidate the actual contributions/claims made.

Well... AMD can just cite their presentation as prior art, and it will be struck down. If anything, Intel was foolish to waste money on bored lawyers.

They can only do it if it was not publicly disclosed. You cannot patent something which has been kept a trade secret.

I don’t understand what you mean. Does AMD’s public disclosure of the microarchitecture prevent them from filing prior art?

Yeah that comment is confusingly worded. I'm not a lawyer, but I do know that there is no explicit relationship between patent-ability and something being a trade secret. The important thing is just that patent applications are public, so once you file, your "secrets" are no longer secret.

This. The whole idea of a patent is that companies share the details of the invention in the patent in exchange for a temporary monopoly on the idea.

I think it's actually the opposite. In the US, the regulations changed several years ago from "First to Invent" (which is ambiguous) to "First to File" (which is pretty clear). In any event, I think you can only claim something as prior art if it's in the public domain. So, if you're very shady/greedy/lack an ethical backbone, and you get your hands on a private brief that isn't protected intellectual property, technically you can file that patent. I assume the owner could try to prove theft if they can determine how the IP was leaked and stolen intently.

Those AMD presentations are publicly available though. It looks like hot chips is a tech conference, and those AMD presentations were published as part of HC 28. I’ll follow up with a link if I can find one, but I think it’s pretty clear they are not in any way trade secrets.

That's the way it should work, at least. But I have a nagging feeling that Intel wouldn't have done this so brazenly if (with the correct amount of lawyers and friendly judges) it wouldn't have seen a chance of the patent being upheld. Then again, "never attribute to malice that which is adequately explained by stupidity", so, who knows?!

Strictly speaking, only the claims matter, and so the text if those would need to be read and used to determine whether it can be invalidated due to prior art. The claims have very little to do with the diagrams supplied, rather they are about a way to compress cache coherency transfers if the line is all zeros (kind of like LPDDR has, in a way).

That said, copying so many of a competitors' slides into figures is still stupid as it is highly unlikely to give a judge a great first impression.

IDK about intel, but quiet a few companies in their sphere offer employees bonuses for patents.

It would not surprise me in the least if this was something some employee did to get a pay bump.

I was thinking about that too - kudos to them if they pulled this off, but I hope they're not with the company anymore...

Not how it works, actually. The system changed in 2013 and now whoever is first to file gets the patent, even if someone else can demonstrate prior art.

https://www.justia.com/intellectual-property/patents/first-t...

AMD should have filed their own patents before talking about it publicly - that's how the game is now played.

If you don't - it's now possible for someone else to file a patent on your own invention.

This is not true - prior art is still a test for striking down a patent in court, it is just not taken into account when two entities are competing to file the same patent at the same time.

They can, but if it is based on industrial espionage, outright theft or plagiarism then that would definitely be a reason for them to lose the patent.

> Not how it works, actually. The system changed in 2013 and now whoever is first to file gets the patent, even if someone else can demonstrate prior art.

This is not true, and it's a really damaging misrepresentation.

First to file deals with interference between parallel applications, it doesn't change prior art based on publication.

If two parties show up claiming patents on the same thing based on unpublished work, under the prior rules the party that was willing to fabricate the earliest date of invention won, under current rules the first to file wins.

The misinformation you're spreading is particularly pernicious because the change increased the incentives for publishing your work early and often (to establish prior art ASAP)-- but the misinterpretation implies you should avoid publishing at all costs (to avoid a third party dishonestly patenting your publications).

The presentation was cited as prior art by Intel.

https://twitter.com/KarbinC/status/1511715859789717510?s=20&...

That’s pretty standard text in tech patent.

What exactly would it mean to file a patent “as a joke?”

No. Those figures are pretty "standard". Neither AMD or Intel have "invented" an OOO processor. Those figures mean nothing. Are the standard "big" boxes in any modern processor. The key details are inside.

E.g., https://uspto.report/patent/grant/10,282,296

But it was accepted?

It wouldn't be the first time a patent for something absurd was granted.

It can't be worse than patenting laser as cat exercise tool. This is not a joke.

It doesn't work that way.

It's one thing to cite it as prior art, but if they are listing it as an example embodiment of their invention then the patent is invalid.

The idea of filing a patent application as a "joke" is nonsense.

Reply on news.ycombinator.com