[00:00:00] Speaker 03:
The first case for argument this morning is 24-2204, Zentian v. Apple. Mr. Rizzoni, whenever you're ready.

[00:00:09] Speaker 04:
Thank you, Your Honors. Kayvon Neruzzi for Appellant, Zentian, and may it please the Court. Your Honors, in this appeal from the underlying decision of the Patent Trial and Appeal Board, The issue is the errors with the board's finding as to the motivation to combine. And the errors are found at two levels. First of all, there was extensive evidence that the combination would defeat Thielen's core purpose of operating as a real-time speech recognition system.

[00:00:44] Speaker 04:
And that evidence had to be addressed by the court in terms of whether it would dissuade from the motivation to combine, including specifically whether the real-time operation of Thielen would be defeated by the combination, as patent owner had extensively shown. The board failed to address that, and that is error under this court's precedence in Matronic versus Teleflex, Polaris versus Arctic Cat, and Intel versus Qualcomm, all of which are cited and discussed in detail in our briefing.

[00:01:16] Speaker 04:
At the second level, Irrespective of that issue, the Board's finding of a motivation to combine is simply not supported by substantial evidence. The Board did not give due weight to the evidence that there would have been a need for substantial reprogramming to allow Thielen to move from its original design to the design that would be required by the combination. Instead, the Board focused on the question of whether the ordinary artisan could have made the necessary programming modifications, but whether the artisan could have made the modifications does not address the separate question of whether the artisan would have been motivated to make the combination in view of all of the substantial programming that would have been required.

[00:02:00] Speaker 02:
Does the claim at issue involve a requirement of real-time Your Honor, it does not, and yet that is... Why does that not matter for deciding how to construct the motivation to combine?

[00:02:23] Speaker 04:
It does matter for deciding how to construct... the motivation to combine, and yet it does not matter for purposes of deciding whether the motivation to combine has to be true to a core purpose of a prior art reference relied on. And so to unpack that a little bit, we can look at this court's three decisions in Medtronic, Polaris, and Intel. And I think Polaris is the easiest one to look at because there, the board, followed a very similar path to here.

[00:02:59] Speaker 04:
It looked to the record and it said, okay, I see that there would have been a benefit from this combination, which would have been a distribution of weight, improvement in the distribution of weight in the prior vehicle. And so that's enough. That's a motivation. And this court said, wait, but the board did not consider the fact that that same prior art reference says you need to keep a low center of gravity and you need to maintain stability. That's a core purpose of that Denny prior art reference.

[00:03:30] Speaker 04:
It's not in the claims. It's what Denny, the prior art, says. And this court said that it's undisputed that the combination would make that vehicle of the combination less stable. And this is contrary to Denny's purpose. And the board did not give consideration or weight to that evidence in making its finding of a motivation to combine. And so this court said that was error.

[00:03:53] Speaker 02:
Don't we have at least some other cases that, I don't know, point a little bit at least in the other direction where you don't, at least as to all of the features of the starting prior art that that piece of prior art says are, are important to what that piece of prior art is doing. If some of those features are not part of the claim, then an artisan might not be not care about shedding some of those features?

[00:04:27] Speaker 02:
Why is this? I think we have such cases probably cited here.

[00:04:34] Speaker 02:
Why is this not more like that?

[00:04:37] Speaker 04:
Yes, Your Honor. So the distinction is at several levels. First of all, there's a difference between features that are discussed in the prior art writ large versus features that are discussed as being core or central to the purpose of the prior art. Second of all, the requirement is not, the rule is not that motivation can never be found simply because it would affect a key purpose of the prior art reference. Rather, it's that the board must affirmatively consider and address whether and why a person of skill would be motivated to make the combination if it would destroy or affect or hinder the core purpose that the prior art identifies.

[00:05:19] Speaker 04:
And that's what didn't happen here. Third of all, though, I think it's very important to note that in this particular proceeding, under the facts of this proceeding, Apple's own expert testified that Thielen's primary purpose is dictation. And without real-time recognition, a dictation system is unusable. And in addition, Apple's own petition, as recognized by the board, stated that real-time speech processing is important to Thielen, and Apple's petition set real-time recognition as a goal and requirement of the combination.

[00:05:53] Speaker 04:
And so that's, for example, at Appendix 235 to 36 and Appendix 954. These are places where Apple's petition emphasized the importance of real-time recognition to the combination. And Appendix 22, the board noted the importance of real-time recognition to the combination. And so... given the centrality of real-time recognition to the purpose of Thielen, what is Thielen for? It's very understandable. You're speaking, and it's doing real-time dictation. Now, if you were to modify that system in a way that it can't achieve that purpose anymore, then it basically is no longer Thielen at all.

[00:06:30] Speaker 04:
It can't do what Thielen set out to do. And under this court's precedence, and the evidence in this case, that raises the necessary question of whether and why a person of ordinary skill would be motivated to make a combination that defeats the core purpose of the primary reference.

[00:06:48] Speaker 00:
That question... It appears that you're really asking us to reweigh the evidence here, which is not something we would do. Why are you not asking for that fundamentally in terms of what you're asking this court to do?

[00:06:59] Speaker 04:
Right, Your Honor. We are not asking you to reweigh the evidence with respect to the real-time recognition issue. We are pointing out that the board failed to do its job, the majority, by the way, right, because there's a vigorous dissent here. So the majority of the board failed to do its job of giving due consideration to the evidence and the issue of the impact of the combination on Thielen's real-time recognition. recognition capability, and whether and why a motivation would have existed in view of that evidence.

[00:07:31] Speaker 04:
So I agree with you that it's not this Court's duty to reweigh the evidence. At the same time, with respect to some other aspects of our arguments, it is this Court's purview to consider whether the Board's findings are supported by substantial evidence, and that does, of course, require consideration of the evidence of record.

[00:07:55] Speaker 04:
So, Your Honors, on the basis of the real-time recognition issue alone, and this court's precedence, the board's decision should be at a minimum vacated. And we think, in fact, it should be reversed because there is no contrary evidence in the record that establishes that the combination could operate in real time, would operate in real time, or that a person of ordinary skill would find the combination to be motivated if the combination could not operate in real time.

[00:08:30] Speaker 04:
So those sort of predicate showings.

[00:08:33] Speaker 02:
Remind me, what did the board say about this real time feature? Did it say any of the three things that you just recited?

[00:08:44] Speaker 04:
It did not say any of those. Or the opposite of the three. Right. It did not say the opposite of the three things, Your Honor. What the board did say is the following, some of which is even more evidence in favor of reversal. The board said that the question of whether or not the combination could operate in real time would have been relevant to a person of ordinary steel in the art in assessing the combination.

[00:09:09] Speaker 00:
An appendix page so that we can review it.

[00:09:11] Speaker 04:
Yes, Appendix 35, Your Honors.

[00:09:15] Speaker 04:
And so in the context of addressing reasonable likelihood of success, the board acknowledged the importance of the real-time question. And yet in the context of its motivation to combine— Reasonable expectation of success? Correct, Your Honors. So there was a separate dispute about reasonable expectation of success, which we do not relitigate in this appeal.

[00:09:45] Speaker 02:
Where on that page is the real-time reference that you're talking about? You said 35. Last paragraph.

[00:09:52] Speaker 03:
The last paragraph. Second paragraph.

[00:10:03] Speaker 04:
It's at the bottom of page 35. While the ability or inability of Apple's proposed combination to have achieved real-time performance may be relevant to the motivation to combine. And then it goes on to say, under the law, the board did not believe it was critical to reasonable expectation of success. But in the context of motivation to combine, the board did not make any finding that the combination would have or could have operated in real time. It simply said that a person of skill in the art, without any particular evidence, simply based on the knowledge and capabilities imbued upon this hypothetical person, would have known how to, quote, minimize the effects of the memory collisions that would be caused by the combination.

[00:10:48] Speaker 00:
What page are you on when you just stated all that?

[00:10:51] Speaker 04:
Yes, Your Honor. I believe that comes up at page 28 of the appendix.

[00:11:00] Speaker 04:
And the board there said a skilled artisan would have expected to be able to minimize the effect of memory collisions based on Chen's teachings and the background knowledge in the art. Now, Chen doesn't teach anything at all about speech recognition or this particular speech recognition combination that requires a very detailed and highly complex set of modifications that start with Thielen, and then you have to take all of Thielen's language models, break them apart, put them into different processors, link them up to a shared cluster memory across multiple different memories.

[00:11:35] Speaker 04:
Chen has nothing to say about that. So it has no solutions to the problems that our expert, Dr. Anderson, identified with a combination with respect to real-time and memory collisions. And certainly, the background knowledge in the art does not extend to using a reference like Chen in this way, which is not a speech recognition reference at all. The person of ordinary seal in the art is in the field of speech recognition, according to the definition Apple itself put forth, not in the field of modifying highly complex parallel.

[00:12:07] Speaker 02:
Was your evidence about the impediment to real-time processing, let's call it, evidence about Chen causing those impediments or even the transition from Thielen to Bailey causing?

[00:12:25] Speaker 04:
It is the transition from the Thielen combination to the Thielen-Bailey-Chen combination. I'm sorry, from the Thielen base reference.

[00:12:34] Speaker 02:
But that's like a two-step process. And tell me if I'm just misremembering. I thought that the board indicated that you didn't dispute a motivation to combine Thielen with Bailey, which means you've already, what, gotten away from the single memory of Thielen. And if that would cause...

[00:13:00] Speaker 02:
a loss or an impediment to real-time processing, then that's an unobjected-to motivation, and you're already at the diminished real-time. And so what do you do with that?

[00:13:14] Speaker 04:
That's not where the problem arises. The problem arises when they go to Chen. And the reason is this. Chen is the one that gives this clustered architecture where you have multiple processors in one cluster linked up to one memory. And that memory, in their combination, only holds part of all of the lexical data. And then you have four different clusters. And so you have all the lexical data broken apart into these different cluster memories. And then in each cluster, multiple processors are going to bombard that memory to get the same, very similar language model data all from the same place.

[00:13:48] Speaker 04:
Is that not true in Thielen? That is not true in Thielen.

[00:13:51] Speaker 02:
Why is that not true in Thielen?

[00:13:53] Speaker 04:
In Thielen, there is one central storage, 340. But aren't all the processors looking at that single memory?

[00:14:02] Speaker 04:
So there's no dispute that Thielen achieves real-time recognition without any problem.

[00:14:07] Speaker 02:
So this problem... I guess I'm just trying to understand in a simple-minded way. Okay.

[00:14:16] Speaker 02:
Is the collision problem not a problem when you have multiple processors trying to use a single memory?

[00:14:24] Speaker 04:
It is not necessarily a problem in the manner of the combination just because there are multiple processors in one memory. It is a problem that comes up specifically because of the way the combination was created. And Dr. Anderson goes into detail about that. For instance, if hypothetically they had set up the combination where the processors that address a particular language model are spread across the different clusters, and so there's actually a distinction between them and which processors are accessing which memory at the same time for a given model, that could theoretically ameliorate the collision load or the collision problem.

[00:15:06] Speaker 04:
But that's not what they did. They created a combination that says... Let's have all the processors that need this information go to the same memory at the same time in a way that is distinct from Thielen. Thielen has it already set up and organized so that each processor can get what it needs from the memory because they're broken apart into different recognition models within that larger memory. In the combination, there are no solutions.

[00:15:31] Speaker 02:
Multiple processors with a single memory in the clusters.

[00:15:36] Speaker 04:
There's no cluster, per se, in theolin.

[00:15:39] Speaker 02:
No, no, no, no, no, in the combination.

[00:15:41] Speaker 04:
Yes, Your Honor. So this is very fact-specific, and this is addressed in detail by Dr. Anderson as to why the problems in this combination don't arise simply because you're trying to use multiple clusters and memories or something generic like that. They arise because of the way –

[00:16:09] Speaker 00:
than your expert in terms of these sort of fact issues?

[00:16:13] Speaker 04:
Your Honor, respectfully, no, because Apple's expert did not address these memory collision problems or provide any solutions to the real-time issue that Dr. Anderson put forth in his original declaration and did not provide any meaningful solutions in his reply declaration. And so there are on one side, and I'm really not oversimplifying, on one side from Apple's side, there are assertions and allegations and assumptions and assurances. And on our side, there was very detailed concrete evidence.

[00:16:45] Speaker 04:
And so at the core, this case is about whether the board may credit allegations and assumptions in the face of extensive detailed evidence to the contrary without a detailed analysis as to why.

[00:17:00] Speaker 03:
Your time's up. Why don't we hear from the other side, and we'll restore a couple minutes. Thank you, Your Honors.

[00:17:23] Speaker 01:
Mr. Lloyd, good morning.

[00:17:25] Speaker 01:
May it please the Court, Seth Lloyd for Apple. The board gave multiple well-supported and independent rationales for finding a motivation to combine. Zentian cannot show error in any of them, let alone all of them, and so the court should affirm. I think given the discussion that we've been having, I'll just jump straight to this issue about real-time processing. I think the first and simplest response to much of what we heard this morning is that the board directly addressed this issue in detail across several pages.

[00:17:56] Speaker 01:
And let me just walk the court through that briefly. So at Appendix 22, the bottom of Appendix 22, the board recognized that real-time processing was one of several motivations that we alleged would have led a person of skill in the art to pursue this combination. Then at Appendix 23...

[00:18:15] Speaker 01:
The board recognized Zentian was arguing against that based on this memory collision argument that you just heard and cited the Zentian expert declaration that was discussed, cited the reference, the Matthew reference that Zentian says the board never addressed. And then the board spent the next several pages from that. Appendix 23, all the way to Appendix 27, discussing the evidence and arguments on both sides about this issue, about whether there would have been a processing slowdown due to memory collisions.

[00:18:46] Speaker 01:
And then the board made two key findings on this issue, directly rejecting Zentian's arguments.

[00:18:53] Speaker 01:
The board found at the break from bottom of 27 over to the top of 28 that any alleged slowdown would not have dissuaded a person of skill in the art from pursuing this combination because of the other benefits Apple found. That's the first paragraph on Appendix 28.

[00:19:12] Speaker 01:
Even if there is this slowdown, the board said, look, a person of skill in the art would have recognized his advantages and disadvantages and would have still pursued this combination. And then, bottom of 28, carrying over to 29, the board, in fact, found that there would not have been a slowdown. The board credited our expert and found Dr. Anderson's testimony, the testimony that you just heard, not credible on this point. So this narrative that the board...

[00:19:38] Speaker 02:
On that last point, which I realize is presented as an indie alternative point, can you explain in a little bit of concrete detail why the collision problem would not be a problem? Yes, and some of it- You're going to have to guess at what's in my head as an image. So go ahead.

[00:20:02] Speaker 01:
I'll do my best to guess. I think some of it goes to some of the colloquy that the court already had, which is that this problem of memory collisions was already a well-known problem in the art because the art was teaching to use shared memory, specifically in the context of speech recognition. That Bailey teaches both a shared memory approach and a private memory approach. So this was not some new problem.

[00:20:26] Speaker 02:
3A and 3B, the figures in Bailey?

[00:20:29] Speaker 01:
Exactly, yes. That's exactly right. And the board, when it walked through the evidence, it noted that this was a known problem. And not only that, but Chen, the reference that we're using for the clustered processing Chen specifically addresses in column five of Chen, kind of the background, it says, look, when you have shared memory, there's going to be this tradeoff where it could lead to slowdown in processing. We're going to teach you a way to avoid that, to kind of find the solution.

[00:21:00] Speaker 01:
optimal point where the benefits of processing and memory meet. And then the board cited columns 14 to 18 of Chen, five columns of Chen that talk about how to deal with this memory problem, this problem of multiple processors accessing the same memory. And so There was ample evidence before the board to support that finding. There's a direct finding on this point. So this narrative, I just respectfully disagree with my friend on the other side that there's this narrative that the board did not address this issue.

[00:21:33] Speaker 01:
And I think for that reason, there was some discussion about, you know, is it even a requirement to consider real-time processing and whether it was a primary purpose of Thalen. I disagree with my colleague about...

[00:21:48] Speaker 01:
some of the legal standard he was articulating, but this court does not even need to reach that here because, in fact, the board did address this issue and found that it would not have dissuaded a person of skill in the art. All of that, as some of the other discussion highlighted, this is extremely fact-specific and fact-intensive analysis, clustered processing.

[00:22:09] Speaker 01:
and the board grappled with all of this, not just the majority, but as the courts saw, there's a dissent here too. So all of this was very well aired in front of the board. The board did the right analysis. The board made findings, crediting our expert and finding their expert not credible on the key points, and that's alone enough for this court to affirm.

[00:22:30] Speaker 00:
I do also... What is your response to opposing counsel's answer to my question that I had previously raised where he said, His expert spoke to it, but your expert was silent on it, and I think he is trying to use that to say there was a lack of substantial evidence support.

[00:22:47] Speaker 01:
Well, so one part of that response, Judge Cunningham, is that regardless of what our experts said, there was prior evidence before the board, and some of which the board cited and which we just talked. So Bailey has a shared memory and must deal with this issue, and Chen specifically in column 5 and then 14 through 18. So there's prior evidence already before Cunningham. But it's also just not true that our expert said or that our expert didn't address this.

[00:23:18] Speaker 01:
It came up in his deposition, and our expert also talked about this issue because it's a well-known problem in the art. And so, again, there was evidence on both sides. The board made a fact-finding, and under the substantial evidence standard, This court defers to the board on those findings. There is also, which the other side didn't address today but did in their briefing, there's an entirely independent rationale that the board adopted here, which Zentian would be required to overcome to get any relief and cannot, which is the board's known problem, known solution alternative finding.

[00:23:51] Speaker 01:
This case, as the board cited to Intel against PACT, this case is that the facts are very similar to Intel against PACT. That case also involved multiprocessor architecture and how to choose the memory, which memory structure to choose for a multiprocessor architecture. And there, one prior reference used one approach to memory, and the proposal was to substitute it with a different memory approach. The board in that case had said, well, rejected the obviousness challenge because it said the challenger had never shown it would have been an improvement to replace the first memory approach with the second.

[00:24:29] Speaker 02:
And this court reversed and said it... The world is not binary between improvement and no motivation. There can be Improvement in some, or the opposite of improvement in one respect, as long as there is some other reason why you would be willing to live with that because you get something else for it. And I guess I took the gist of the argument on the other side, focused on real time, is that, at least in Thielen, There's no point in doing anything if you have to sacrifice the real time.

[00:25:08] Speaker 02:
That's what I think your friend called a core aspect of what Thielen teaches. Can you address that slightly more specifically?

[00:25:18] Speaker 01:
Yes. So my first response is, and I won't repeat it all, but my first response is the board considered that on the facts and just disagreed that those are the facts here. But I think the second response is on the legal side, which this court has recognized.

[00:25:36] Speaker 01:
and it flows directly from KSR, that it's improper to consider a prior art only for its primary purpose. That was one of the things that the Supreme Court reversed in KSR, that a person of skill in the art may recognize that what a prior art reference teaches can be applied in other ways beyond its primary purpose. And then in addition, and this court's decision... But we are really, I think...

[00:25:59] Speaker 02:
single-mindedly focused on the motivation question, not on the teaching question. And if indeed somebody would, or no relevant artisan would be motivated to change a particular piece of prior art, it doesn't matter if it teaches a bunch of other things.

[00:26:17] Speaker 01:
Well, so maybe I'm not articulating as well as I want to, but I think what KSR says and what decisions like Axonics from this court have held is that it would be error to confine the motivation inquiry to only a purpose that's disclosed in the prior art and that is not shared with the claimed invention. So here the claimed invention has no requirement for real-time processing. So what

[00:26:47] Speaker 02:
But then in axonics, the definition of the person of skill in the art was somebody looking at a different part of the brain from the one that was taught in the primary reference. And so you couldn't possibly say that person would be looking only for improvements in the wrong part of the brain.

[00:27:08] Speaker 02:
What here is the evidence of a motivation to make this combination, if it were true, and I realize you pointed to where the board found that it was not true, that you couldn't really do real-time speech processing?

[00:27:24] Speaker 01:
Yeah, so the definition of a person of skill in the art, which is unrebutted here, is just speech processing generally. And there are other applications of speech processing that are not real-time. Our expert testified to that, and I can give you the exact site for that.

[00:27:43] Speaker 02:
What would be a homey example? Just an ordinary example where you want speech recognition, but you don't care if it comes in the mail the next day.

[00:27:53] Speaker 01:
Yeah, so I don't know if it's delayed by a day, but for example, dictation, where you're going to, where, for example, you might be recording a bunch of your thoughts, and then later on you're going to look back at it. It's okay if the system takes longer to do that than a type of system where maybe you're sending text messages that you want sent very quickly. So there are different applications. That is in the record. And in fact, I think one other key point to this whole argument is that Thalen itself expressly recognizes in its background that Let me just make sure I find the right site here.

[00:28:34] Speaker 01:
So Thalen is in column 6 of Thalen at 2231, lines 40 to 45. It expressly recognizes that there are other applications for its system beyond real-time. It just says that real-time is one preferable application.

[00:28:48] Speaker 01:
approach. And so Thalen is recognizing just exactly what I was just discussing with Judge Toronto, that there are other applications. Yes, this is one preferable one, but it applies more broadly. And so even kind of that narrative of this is the sole purpose of Thalen is not present here.

[00:29:08] Speaker 01:
Unless the court has any further questions, we'd ask that you affirm.

[00:29:31] Speaker 04:
Your Honors, the last point I'd like to address first. Apple admitted at the beginning of this argument, as well as numerous times below, that their own combination requires the real-time processing, and that's stated at Appendix 22. It's also stated throughout their petition and expert declaration. So whether Thielen could be used for other purposes still does not address not either Thielen's core purpose or the way that the combination itself was set forth and described.

[00:30:04] Speaker 04:
In addition, as for what was cited at Appendix 28-29 regarding the real-time issue and the Board's consideration, there the Board addressed a strawman argument. What the Board said at 29 is that a person of ordinary skill was not shown to have been unable to increase the number of processors within a cluster without resulting in a slowdown with respect to that cluster or the system as a whole. But that very generic question is not the issue presented by our evidence.

[00:30:36] Speaker 04:
Our evidence has to do with this specific combination between Thiel and Bailey and Chen, and when you set it up the way that Apple said, which is described in detail at Appendix 58 by the dissent, then you have a problem with collisions that causes the failure of real time. And the board at Appendix 28 to 29 did not address that specific evidence. And that's the shortcoming. Now, as to the known problem, known solution doctrine, this is not such a case. because even under Intel versus PACT-XPP, the combination must be found to be suitable.

[00:31:12] Speaker 04:
And the entire argument here is that the failure of real-time makes the combination unsuitable, and the board did not adequately consider that. In addition, PACT-XPP, Intel versus PACT, requires a known technique. Chen is not even in the field of speech recognition, and we've detailed numerous other distinctions between Intel and this case.

[00:31:33] Speaker 04:
Thank you, Your Honor.

[00:31:34] Speaker 03:
We thank both sides. The case is submitted.