The California Institute v. Broadcom Limited

• Court: Court of Appeals for the Federal Circuit
• Decided: February 4, 2022
• Docket: 20-2222
• Status: Published

Opinion

Case: 20-2222 Document: 63 Page: 1 Filed: 02/04/2022

United States Court of Appeals for the Federal Circuit ______________________

CALIFORNIA INSTITUTE OF TECHNOLOGY, Plaintiff-Appellee

v.

BROADCOM LIMITED, NKA BROADCOM INC., BROADCOM CORPORATION, AVAGO TECHNOLOGIES LIMITED, NKA AVAGO TECHNOLOGIES INTERNATIONAL SALES PTE. LIMITED, APPLE INC., Defendants-Appellants ______________________

2020-2222, 2021-1527 ______________________

Appeals from the United States District Court for the Central District of California in No. 2:16-cv-03714-GW- AGR, Judge George H. Wu. ______________________

Decided: February 4, 2022 ______________________

KATHLEEN M. SULLIVAN, Quinn Emanuel Urquhart & Sullivan, LLP, Los Angeles, CA, argued for plaintiff-appel- lee. Also represented by JAMES R. ASPERGER; BRIAN P. BIDDINGER, EDWARD J. DEFRANCO, New York, NY; TODD MICHAEL BRIGGS, KEVIN P.B. JOHNSON, Redwood Shores, CA; DEREK L. SHAFFER, Washington, DC; KEVIN ALEXANDER SMITH, San Francisco, CA. Case: 20-2222 Document: 63 Page: 2 Filed: 02/04/2022

WILLIAM F. LEE, Wilmer Cutler Pickering Hale and Dorr LLP, Boston, MA, argued for defendants-appellants. Also represented by LAUREN B. FLETCHER, MADELEINE C. LAUPHEIMER, JOSEPH J. MUELLER; STEVEN JARED HORN, DAVID P. YIN, Washington, DC; MARK D. SELWYN, Palo Alto, CA. ______________________

Before LOURIE, LINN, and DYK, Circuit Judges. Opinion for the Court filed by Circuit Judge LINN. Opinion concurring-in-part and dissenting-in-part filed by Circuit Judge DYK. LINN, Circuit Judge. Broadcom Limited, Broadcom Corporation, and Avago Technologies Ltd. (collectively “Broadcom”) and Apple Inc. (“Apple”) appeal from the adverse decision of the District Court for the Central District of California in an infringe- ment suit filed by the California Institute of Technology (“Caltech”) for infringement of its U.S. Patents No. 7,116,710 (“the ’710 patent”), No. 7,421,032 (“the ’032 pa- tent”), and No. 7,916,781 (“the ’781 patent”). Because the district court did not err in its construction of the claim limitation “repeat” and because substantial ev- idence supports the jury’s verdict of infringement of the as- serted claims of the ’710 and ’032 patents, we affirm the district court’s denial of JMOL on infringement thereof. We also affirm the district court’s conclusion that claim 13 of the ’781 patent is patent-eligible but vacate the jury’s verdict of infringement thereof because of the district court’s failure to instruct the jury on the construction of the claim term “variable number of subsets.” We thus remand for a new trial on infringement of claim 13 of the ’781 pa- tent. We further affirm the district court’s summary judg- ment findings of no invalidity based on IPR estoppel and its determination of no inequitable conduct. We affirm the Case: 20-2222 Document: 63 Page: 3 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 3

district court’s decision with respect to its jury instructions on extraterritoriality. But because Caltech’s two-tier dam- ages theory cannot be supported on this record, we vacate the jury’s damages award and remand for a new trial on damages. BACKGROUND I. The Caltech Patents Caltech’s ’710 and ’032 patents disclose circuits that generate and receive irregular repeat and accumulate (“IRA”) codes, a type of error correction code designed to improve the speed and reliability of data transmissions. Wireless data transmissions are ordinarily susceptible to corruption arising from noise or other forms of interfer- ence. IRA codes help to identify and correct corruption af- ter it occurs. The encoding process begins with the processing of data before it is transmitted. The data consists of infor- mation bits in the form of 1’s and 0’s. The information bits are input into an encoder, a device that generates code- words comprised of parity bits and the original information bits. Parity bits are appended at the end of a codeword. Codewords are created in part by repeating information bits in order to increase the transmission’s reliability. When noise or other forms of interference introduce errors into the codewords during transmission, the decoder iden- tifies these errors and relies on the codeword’s redundant incorporation of the original string of information bits to correct and eliminate the errors. Before Caltech’s patents, error correction codes had al- ready incorporated repetition and irregular repetition. These codes, however, were less than optimally efficient be- cause they were either encoded or decoded in quadratic time, which meant that the number of computations Case: 20-2222 Document: 63 Page: 4 Filed: 02/04/2022

required to correct a given number of bits far exceeded the number of bits ultimately corrected. In the ’710 and ’032 patents, the IRA codes are linear- time encodable and decodable, rather than quadratic. ’710 patent, col. 2, ll. 6–7 (“The encoded data output from the inner coder may be transmitted on a channel and decoded in linear time.”); id. col. 2, l. 59 (“The inner coder 206 may be a linear rate-1 coder.”); id. col. 3, ll. 25–26 (“An IRA code is a linear code.”). Using a linear code means that the re- lationship between the bits corrected and the computations required is directly proportional. Minimizing the number of calculations that an encoder or decoder must perform permits smaller, more efficient chips with lower power re- quirements. The claimed improvement involves encoding the infor- mation bits through a process of irregular repetition, scrambling, summing, and accumulation. Repeating in- putted information bits is necessary to increase the relia- bility of data transmissions, and irregular repetition minimizes the number of times that information bits are repeated. Minimizing the number of times that an infor- mation bit is repeated is crucial to the efficiency of the claimed inventions because the repetitions impact the de- vice’s coding rate or speed, as well as the code’s complexity. The fewer repeated bits there are, the fewer number of computations that an encoder must perform, which in turn permits smaller circuits, decreased power requirements, and decreased operating temperatures in devices incorpo- rating the circuits. The claims and accompanying specifications of the Cal- tech patents make clear that each inputted information bit must be repeated. The parties agree that every claim at issue requires irregular repetition of information bits ei- ther explicitly or via the court’s construction. This is so even where the irregular repetition is not expressly Case: 20-2222 Document: 63 Page: 5 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 5

required by the claims. For example, the agreed-upon con- struction of a Tanner graph in the ’032 patent requires that “every message bit is repeated . . . .” J. App’x 33. Further- more, the claims and accompanying specifications make clear that each bit must be repeated irregularly, stating, for example in the ’710 patent, “a fraction of the bits in the block may be repeated two times, a fraction of bits may be repeated three times, and the remainder of bits may be re- peated four times.” ’710 patent, col. 2, ll. 53–58. The ‘781 patent discloses and claims a method for cre- ating codewords in which “information bits appear in a var- iable number of subsets.” Before trial, Apple and Broadcom sought summary judgment that claim 13 was unpatentable under 35 U.S.C. § 101. After finding that the claims were directed to a patent-eligible subject matter (step 1 of Alice 1)—a method of performing error correction and detection encoding with the requirement of irregular repetition—the court declined to reach whether they con- tained an inventive concept (step 2 of Alice). To support patentability, Caltech argued that the “variable number of subsets” language required irregular information bit repe- tition. The district court agreed and adopted and relied on Caltech’s interpretation to deny summary judgment of un- patentability. No party on appeal challenges this claim in- terpretation. II. The Accused Products Caltech alleged infringement by certain Broadcom Wi- Fi chips and Apple products incorporating those chips, in- cluding smartphones, tablets, and computers. The accused Broadcom chips were developed and supplied to Apple pur- suant to Master Development and Supply Agreements

1 Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208 (2014). Case: 20-2222 Document: 63 Page: 6 Filed: 02/04/2022

negotiated and entered into in the United States. Caltech specifically identified as infringing products two encode1·s contained in the Broadcom chips-a Richardson-Urbanke ("RU") encoder and a low-area ("LA") encoder. In the ac- cused encoders, incoming information bits are provided to AND gates in the RU encoder or multiplexers in the LA encodei·. Throughout the ti·ial and on appeal, the parties treated AND gates and multiplexers as functionally identical for all relevant issues. It thus suffices to describe in detail the RU encoder only. In the RU encoder, each information bit is simultaneously fed as one input to 972 separate AND gates. Each AND gate receives a second input-a "parity- check" or "enable" bit of 0 or I-derived from a low-density parity check matrix. This matrix is an anay of l 's and O's. A low-density parity check matrix is one in which the num- ber of l's in the matrix is significantly fewer than the num- ber of O's. In its brief, Broadcom presents the following table, us- ing the example of the functioning of a single AND gate, to show how outputs ai·e determined by the two inputs:

Input 1 Input 2 AND Gate (Information Bit) (Parity-Check Bit) Output 0 0 0 0 1 0 1 0 0 1 1 1

For each AND gate, the output of the gate is 1 if both inputs (the information bit and the parity-check bit) are l ; otherwise, the output is 0. One consequence of this logic is that if the parity-check bit is 1 (as shown in rows two and four) , then the output is identical to the information-bit in- put. If the parity-check bit is 0, the output is 0, regardless Case: 20-2222 Document: 63 Page: 7 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 7

of the value of the input (rows one and three). Throughout trial, the parties referred to parity-check bits and enable bits interchangeably. Parity-check bits determine the ac- tion of the AND gates, which are open/on when the parity- check bit is 1 and closed/off when the parity-check bit is 0. Caltech sued Broadcom and Apple on May 26, 2016, al- leging infringement under 35 U.S.C. § 271 by Broadcom wireless chips and Apple products incorporating those chips. Both defendants denied that any of the accused de- vices infringed Caltech’s patents, and in turn asserted counterclaims for declaratory judgment of non-infringe- ment, invalidity under 35 U.S.C. §§ 101, 102, 103, and/or 112, and unenforceability due to inequitable conduct. III. Pre-Trial Proceedings Before trial, Apple filed multiple IPR petitions chal- lenging the validity of the claims at issue, relying on vari- ous prior art references. The Patent Trial and Appeal Board (“PTAB” or “Board”) issued a number of written de- cisions, which concluded that Apple failed to show the chal- lenged claims were unpatentable as obvious. Before the district court, Apple and Broadcom argued that the as- serted claims would have been obvious over new combina- tions of prior art not asserted in the IPR proceedings. The district court granted summary judgment of no in- validity, interpreting 35 U.S.C. § 315(e)(2) as precluding parties from raising invalidity arguments at trial that they reasonably could have raised in their IPR petitions. It also denied the motion filed by Apple and Broadcom for sum- mary judgment of invalidity under 35 U.S.C. § 101 for the ’781 patent. The district court granted Caltech’s summary judgment motion as to inequitable conduct, finding no in- equitable conduct with respect to Caltech’s failure to dis- close Richardson99 during prosecution. The district court reasoned that this prior art reference was not but-for ma- terial to the PTO’s grant of Caltech’s patents. Case: 20-2222 Document: 63 Page: 8 Filed: 02/04/2022

The district court also conducted a Markman hearing and initially construed the claim limitation “repeat.” That construction is germane to all of the asserted claims. At the conclusion of the Markman hearing, the district court construed “repeat” to have its plain and ordinary meaning. The district court noted that the repeated bits “are a con- struct distinct from the original bits from which they are created,” but that they need not be generated by storing new copied bits in memory. IV Trial Proceedings A. Infringement of the ’710 and ’032 Patents At trial, Caltech argued that the accused chips in- fringed claims 20 and 22 of the ’710 patent and claims 11 and 18 of the ’032 patent. Both groups of claims explicitly require irregular repetition; i.e., repetition of groups of in- formation bits an irregular number of times. Claims 20 and 22 of the ’710 patent depend from claim 15, which claims: 15. A coder comprising: a first coder having an in- put configured to receive a stream of bits, said first coder operative to repeat said stream of bits irreg- ularly and scramble the repeated bits; and a second coder operative to further encode bits output from the first coder at a rate within 10% of one. ’710 patent, col. 8, ll. 1–6. Claims 11 and 18 of the ’032 patent cover devices for encoding and decoding pursuant to a Tanner graph: 2

2 During claim construction, the parties agreed that a Tanner graph is a visual representation of the “con- straints that determine the parity bits” created by an IRA code. J. App’x 33. Case: 20-2222 Document: 63 Page: 9 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 9

11. A device comprising: an encoder configured to receive a collection of message bits and encode the message bits to generate a collection of parity bits in accordance with the following Tanner graph:

·········/P\ •········\p): ·~...~··

·········•18r I \c} I ·~....•

•·····---r3\: . . ...\:8L c✓.

18. A device comprising: a message passing de- coder configured to decode a received data stream that includes a collection of parity bits, the message passing decoder comprising two or more check/var- iable nodes operating in parallel to receive mes- sages from neighboring check/variable nodes and send updated messages to the neighboring varia- ble/check nodes, wherein the message passing de- coder is configured to decode the received data stream that has been encoded in accordance with the following Tanner graph: Case: 20-2222 Document: 63 Page: 10 Filed: 02/04/2022

~ •·······/8f I .. . . .\o.L I ... .., ':'

•···· ·•18\~ •···· ··.\bi~ ..,._..:

’032 patent, col. 8, l. 63–col. 9, l. 34; col. 9, l. 57–col. 10, l. 42. The district court’s claim construction ruling required that the Tanner graphs in claims 11 and 18 also perform repetition. J. App’x 33 (defining Tanner graph as a depic- tion of “an IRA code as a set of parity checks where every message bit is repeated, at least two different subsets of message bits are repeated a different number of times”). No party challenges this construction on appeal. During trial, the district court revisited and clarified its earlier claim construction ruling of the term “repeat” and instructed the jury that repeat means “generation of additional bits, where generation can include, for example, duplication or reuse of bits.” Apple and Broadcom then ar- gued that the chips did not infringe the ’710 and ’032 pa- tents because they did not repeat information bits at all, much less irregularly. The jury ultimately found infringe- ment of all the asserted claims. Broadcom and Apple filed post-trial motions for JMOL and a new trial, challenging the jury’s infringement verdict. The district court denied JMOL, finding no error in its claim construction ruling and Case: 20-2222 Document: 63 Page: 11 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 11

concluding that the verdict was supported by substantial evidence. B. Infringement of the ’781 Patent At trial, Caltech also argued that the accused chips in- fringed claim 13 of the ’781 patent. That patent discloses and claims a method for creating codewords in which “in- formation bits appear in a variable number of subsets.” Claim 13 recites: A method of encoding a signal, comprising: receiving a block of data in the signal to be encoded, the block of data including information bits; and performing an encoding operation using the infor- mation bits as an input, the encoding operation in- cluding an accumulation of mod-2 or exclusive-OR sums of bits in subsets of the information bits, the encoding operation generating at least a portion of a codeword, wherein the information bits appear in a variable number of subsets. ’781 patent, col. 8, ll. 7–16. Despite its construction at the summary judgment stage that the claim term “variable number of subsets” re- quires irregular information bit repetition, the district court declined to provide the jury with an instruction of that claim construction determination and the jury deter- mined that Apple and Broadcom infringed claim 13 of the ’781 patent. Broadcom and Apple filed JMOL and new trial motions arguing that the district court erred in refusing their requested instruction and that JMOL of noninfringe- ment was appropriate because the irregular repetition re- quirement was not satisfied. In denying these post-trial motions, the district court concluded that it was “within its Case: 20-2222 Document: 63 Page: 12 Filed: 02/04/2022

discretion” not to issue this instruction so as not to “confuse the record on this issue.” C. Damages To compensate for Broadcom and Apple’s infringement, Caltech proposed a two-tier damages theory, which sought different royalty rates from each of the infringers despite the fact that liability arose from the same accused technol- ogy in the same chips. Even though the district court voiced its discomfort with the two-tier theory, it allowed Caltech to present the theory to the jury, which relied on it to award Caltech $270,241,171 for Broadcom’s infringe- ment and $837,801,178 for Apple’s infringement. The jury’s damages award was based on Caltech’s experts’ tes- timony, admitted over Broadcom and Apple’s objection. Appellants challenged the damages award in their post- trial motions, which the district court denied. The district court entered judgment against Broadcom totaling $288,246,156, and against Apple totaling $885,441,828. These awards included pre-judgment interest, as well as post-judgment interest and an ongoing royalty at the rate set by the jury’s verdict. Broadcom and Apple appeal. We have jurisdiction pur- suant to 28 U.S.C. §1295(a)(1). DISCUSSION I. Standard of Review Claim construction is reviewed de novo when relying on intrinsic evidence. Teva Pharms. USA, Inc. v. Sandoz, Inc., 574 U.S. 318, 333 (2015). Infringement and damages are reviewed for substantial evidence. Lucent Techs., Inc. v. Gateway, Inc., 580 F.3d 1301, 1309, 1324 (Fed. Cir. 2009). Statutory interpretation is reviewed de novo. Power Integrations v. Semiconductor Components Indus., LLC, 926 F.3d 1306, 1313 (Fed. Cir. 2019). Patent-eligibility un- der 35 U.S.C. § 101 is reviewed de novo. Recognicorp, LLC Case: 20-2222 Document: 63 Page: 13 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 13

v. Nintendo Co., 855 F.3d 1322, 1326 (Fed. Cir. 2017). We review patent jury instructions on patent law issues de novo, asking if the instructions were legally erroneous and prejudicial. Bettcher Indus., Inc. v. Bunzl USA, Inc., 661 F.3d 629, 638-39 (Fed. Cir. 2011). We review a district court’s order denying JMOL under the standard applied by the regional circuit. Apple, Inc. v. Samsung Electronics Co., Ltd., 839 F.3d 1034, 1040 (Fed. Cir. 2016). In the Ninth Circuit, JMOL “is proper when the evidence permits only one reasonable conclusion and the conclusion is contrary to that of the jury.” See Monroe v. City of Phoenix, 248 F.3d 851, 861 (9th Cir. 2001). The Ninth Circuit explains that “[t]he evidence must be viewed in the light most favorable to the nonmoving party, and all reasonable inferences must be drawn in favor of that party.” Id. The Ninth Circuit reviews a district court’s de- cision to deny JMOL de novo. Id. II. Infringement A. The ’710 and ’032 Patents Broadcom and Apple argue that the district court erro- neously construed “repeat,” contending that the accused AND gates and multiplexers do not “repeat” information bits in the manner claimed, but instead combine the infor- mation bits with bits from a parity-check matrix to output new bits reflecting that combination. Broadcom and Apple further argue that the AND gates and multiplexers also do not generate bits “irregularly,” asserting that they output the same number of bits for every information bit. Caltech argues in response that expert testimony throughout the record establishes that every information bit is repeated an irregular number of times. According to Caltech, the jury heard testimony explaining that in the RU devices every bit in the stream of information bits is fed by wire simultane- ously to the information inputs of all 972 AND gates and that at any time, at least 3 and up to 12 of those AND gates Case: 20-2222 Document: 63 Page: 14 Filed: 02/04/2022

will be enabled to repeat that bit at the output of the AND gates. We find no error in the district court’s construction of the term “repeat” and agree with Caltech that substan- tial evidence in the record supports the jury’s verdict on in- fringement. 1. Claim Construction of “repeat” The district court construed “repeat” to mean “genera- tion of additional bits, where generation can include, for example, duplication or reuse of bits” (emphasis added). J. App’x 171. Broadcom and Apple argue that that construc- tion is inconsistent with the claim language, the specifica- tion and the construction given by another judge in a different case. 3 Caltech argues in response that the plain claim language requiring repeating information bits does not require generating new, distinct bits and that the dis- trict court was correct in construing the term to not exclude the reuse of bits. We agree with Caltech. The district court correctly observed that the claims re- quire repeating but do not specify how the repeating is to occur: “The claims simply require bits to be repeated, with- out limiting how specifically the duplicate bits are created or stored in the memory.” J. App’x 10. The specifications confirm that construction and describe two embodiments, neither of which require duplication of bits. The district court carefully and fully considered both the language of the claims and that of the written description and faithfully applied our precedent to reach the construction made dur- ing the trial and presented to the jury. We are not

3 Broadcom and Apple misplace reliance on the con- struction of the term “repeat” made on an undeveloped rec- ord in the context of a summary judgment motion. See California Institute of Technology v. Hughes Communica- tions Inc., 35 F. Supp. 3d 1176 (C.D. Cal. 2014). Case: 20-2222 Document: 63 Page: 15 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 15

persuaded that the district court erred in construing the term “repeat” and, therefore, affirm the same. 2. JMOL on Infringement Broadcom and Apple argue that the evidence before the jury on infringement permitted only one verdict, namely no infringement, and that the district court erred in denying JMOL. Broadcom and Apple put forth two rationales for noninfringement of the “irregular repeat” requirement, Ap- pellant’s Br. 27–31. First, looking at each gate alone and commenting on the “repeat” requirement, they argue that the AND gate does not “repeat” the inputted information bit “because the AND gate’s output depends on not only the information bit but also the parity-check-matrix bit.” Ap- pellant’s Br. 29. Second, focusing on the “irregular” half of “irregular repeat,” they argue that “even if the outputted bits could be deemed ‘repeats’ of the information bits,” “any repetition is not ‘irregular’ because each information bit leads to the same number of outputted bits.” Appellant’s Br. 30. Caltech argues in response that the jury was provided with substantial evidence to support the verdict of infringe- ment and that the district court correctly denied JMOL. Caltech asserts that the fact that an AND gate doesn’t have an information-bit/output match for every information bit hardly means that it isn’t repeating any information bit. Appellee’s Br. 21–22 (citing J. App’x 3036–38). All that matters, according to Caltech, is that sometimes there is such a match that qualifies as a “repeat,” so long as each and every bit is repeated at least once. Caltech argues that Broadcom ignores ample expert testimony, which the jury could credit, that sometimes an AND gate repeats an infor- mation bit and that, taking the 972 AND gates together, the carefully designed parity-bit table/matrix meant that “the products output and store information bits between two and twelve times.” Appellee’s Br. 22. Caltech asserts Case: 20-2222 Document: 63 Page: 16 Filed: 02/04/2022

that, considering the system as a whole, each information bit is in fact repeated, and they are not all repeated the same number of times. We agree with Caltech. Caltech’s expert, Dr. Matthew Shoemake began his tes- timony with reference to the exemplary table reproduced above. See J. App’x 3036–38. He explained that in the par- ity-check-bit-equals-1 situation (second and fourth rows of the table), the output bit is a “repeat” of the information- bit input. Where the parity-check bit is 1, the gate affirm- atively enables the information bit to be duplicated as the output bit. That is a “repeat.” That is so, he explained, because the information bit in that situation “flows through” to appear again in the output. He also addressed the one other situation where the output bit is identical to the information bit, namely, in the first row of the above table, where both the information bit and the parity-check bit are 0, and so is the output. Despite the identity of the information bit and the output bit, he explained, that situ- ation does not involve a “repeat.” A 0 parity-check bit turns every information bit (0 or 1) into a 0 output, so the output bit in that situation tells one nothing about the information bit. Since the whole point of this encoding scheme is to use outputs that give information about the information bits, a 0 parity-check bit does not produce a “repeat” even when the information-bit input and the output are the same. Broadcom’s expert, Dr. Wayne Stark, expressly recognized that this was exactly what Dr. Shoemake said in his testi- mony. J. App’x 3956 (“He said it’s a repeat only if the ena- ble [parity-check] signal is a one and it’s not a repeat if an enable [parity-check] symbol is a zero.”). Dr. Shoemake also explained to the jury that “flow through” means that the information bit is repeated at the output gate. See, e.g., J. App’x 2810, 2812, 3017–19. When the information bit “flows through” to the output gate be- cause the parity-check bit is 1, that’s a repeat, both accord- ing to the expert’s usage and a plain understanding of the Case: 20-2222 Document: 63 Page: 17 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 17

word “repeat.” See, e.g., J. App’x 3038. When the infor- mation bit is not allowed to flow through (because the par- ity-check bit is 0), that’s not a repeat (even though both the information bit and the output bit are 0). 4 In explaining the operation of the RU encoder itself, Dr. Shoemake testified that it contains “972 mac_reg modules [AND gates], and the information bits are connected to every single one of them.” J. App’x 2831. He further testi- fied that: “[D]epending on which information bit it is, 3 to 12 of these gates are enabled which then allows 3 to 12 … [information bits] to flow through 3 to 12 times and since that number varies, there’s irregular repetition,” J. App’x 3034-35; “[W]hat really happens in the accused products, the tables tell you how many times should information bit number one be repeated. And the tables I've mentioned several times that they allow information bits, and I should force information bits to be repeated between 3 and 12 times,” J. App’x 3080; and “[T]he information bit starts off in one location in the chip, and then it’s connected to 972 distinct locations so it can be irregularly repeated in this architecture.” J. App’x 3018. Dr. Shoemake’s position was consistent throughout his testimony: the physical connection of the first inputs of all 972 AND gates for simultaneous receipt of the information bit stream and the connection of the parity-bit system to the other inputs of the AND gates to selectively enable 3 to 12 of those gates at any time together implement irregular

4 Caltech’s Red Brief incorrectly cited this example as representing a repeat. Red. Br. 21. This was evidently error, given that it directly contracted the directly cited pages of Dr. Shoemake’s testimony. This error does not, however, change the fact that Caltech correctly identified the substantial trial testimony on which the jury could base its decision. Case: 20-2222 Document: 63 Page: 18 Filed: 02/04/2022

repetition. Dr. Shoemake explained that this is exactly what one sees when one looks at the “overall architecture” (“whole architecture”), not each gate alone. J. App’x 3031, 3035, 3038. As he specifically testified: Q: Your position, your opinion . . . is that that branch wire creates 972 repeat bits within the meaning of the claims in the Caltech patents; cor- rect? A: So based on my analysis, this wire going to the Mac rag modules and the AND gates under control of the tables that are stored in the RU encoder actually allows the infor- mation bits to flow through [a] different number of times. It’s always 3 to 12 times for a particular information bit. And so [i]n my analysis, this is exactly how the RU en- coder is implementing irregular repetition of information bits. J. App’x 3019 (emphasis added). For the foregoing reasons, substantial evidence sup- ports the jury’s verdict of infringement of the ’710 and ’032 patents. We are not persuaded that the record before the jury permits only a verdict of no infringement. We there- fore affirm the district court’s denial of JMOL. B. The ‘781 Patent 1. Patent Eligibility Broadcom and Apple contend that claim 13 is not pa- tent eligible under 35 U.S.C. § 101. Broadcom and Apple’s briefing on this issue was cursory and relied solely on an argument that claim 13 is ineligible because it depends on mathematical operations. Caltech contends that the ’781 patent is directed to a patent-eligible method of performing error correction and detection encoding with the require- ment of irregular repetition. It asserts that the claim Case: 20-2222 Document: 63 Page: 19 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 19

limitation “variable number of subsets” requires irregular information bit repetition. The mere fact that Caltech’s claim employs a mathe- matical formula does not demonstrate that it is patent in- eligible. See Diamond v. Diehr, 450 U.S. 175, 187 (1981) (“[A] claim drawn to subject matter otherwise statutory does not become nonstatutory simply because it uses a mathematical formula, computer program, or digital com- puter.”). Claim 13 does not claim a mathematical formula as such. It claims more than a mathematical formula be- cause it is directed to an efficient, improved method of en- coding data that relies in part on irregular repetition. This alleged improvement is not patent ineligible simply be- cause it employs a mathematical formula. 2. Infringement Broadcom and Apple argue that even if claim 13 is di- rected to patent eligible subject matter, the infringement verdict as to claim 13 cannot stand. As discussed above, the parties agree that claim 13 requires irregular repeti- tion, but dispute whether the district court erred in refus- ing to instruct the jury that the ’781 patent’s “variable number of subsets” limitation requires irregular repetition. The district court’s sole ground for refusing to instruct the jury of the interpretation the parties and the court reached during summary judgment was to avoid “confus[ing] the record on this issue.” J. App’x 207. This was error and requires remand for a new trial on infringement. Sulzer Textil A.G. v. Picanol N.V., 358 F.3d 1356, 1366 (Fed. Cir. 2004) (“[I]t is the duty of trial courts in patent cases in which claim construction rulings on disputed claim terms are made . . . to inform jurors both of the court's claim con- struction rulings on all disputed claim terms and of the ju- ry's obligation to adopt and apply the court's determined meanings[.]”). On remand, the district court must instruct the jury as to the proper construction of the claim limita- tion “variable number of subsets.” Case: 20-2222 Document: 63 Page: 20 Filed: 02/04/2022

III. Validity and IPR Estoppel Apple and Broadcom contend that the district court erred in granting summary judgment of no invalidity, bar- ring them from presenting an invalidity case at trial on the ground of statutory estoppel. In the district court proceed- ings, the parties challenged the patents’ invalidity, relying on grounds the PTAB did not address in its earlier insti- tuted IPR decisions. The district court nonetheless held that these challenges were barred by estoppel because Ap- ple and Broadcom were aware of the prior art references at the time they filed their IPR petitions and reasonably could have raised them in those petitions even if they could not have been raised in the proceedings post-institution. Before the district court, Broadcom and Apple brought counterclaims seeking declaratory judgment of invalidity under § 103. The district court’s summary judgment or- ders disposed of the parties’ affirmative defenses as well as their counterclaims. We therefore consider whether this ruling was erroneous and review the grant of summary judgment de novo. Synopsys, Inc. v. Mentor Graphics Corp., 839 F.3d 1138, 1146 (Fed. Cir. 2016). When IPR proceedings result in a final written deci- sion, 35 U.S.C. § 315(e)(2) precludes petitioners from rais- ing invalidity grounds in a civil action that they “raised or reasonably could have raised during that inter partes re- view.” Shaw Industries Group, Inc. v. Automated Creel Systems, Inc., 817 F.3d 1293, 1300 (Fed. Cir. 2016) (empha- sis added). In Shaw, this court held that IPR “does not begin until it is instituted.” Id. If IPR “does not begin until it is instituted,” grounds raised in a petition (or that rea- sonably could have been raised in a petition) were neces- sarily not raised “during the IPR.” Id. Only the grounds actually at issue in the IPR were raised, or reasonably could have been raised in the IPR. Thus, estoppel did not bar the petitioner in Shaw from presenting a petitioned- Case: 20-2222 Document: 63 Page: 21 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 21

for, non-instituted ground in future proceedings because the petitioner could not reasonably have raised the ground during IPR. Id. Shaw was followed in HP Inc. v. MPHJ Technology Investments, LLC, 817 F.3d 1339, 1347–48 (Fed. Cir. 2016). At the time Shaw was decided, the PTAB often instituted review on less than all the grounds raised in a petition, which left some grounds unadjudicated on the merits. Before Shaw, we had held in Synopsys, Inc v. Men- tor Graphics Corp., 814 F.3d 1309, 1314–15 (Fed. Cir. 2016), that the PTAB’s final decision need not address every claim raised in a petition. Under such circum- stances, we concluded that Congress could not have in- tended to bar later litigation of the issues that the PTAB declined to consider. After Shaw, several district courts concluded that Shaw does not allow a petitioner to avoid estoppel as to all arguments that could have been raised in the petition. See, e.g., SiOnyx, LLC v. Hamamatsu Photonics K.K., 330 F. Supp. 3d 574, 602 (D. Mass. 2018) (determining that estop- pel applies to grounds not included in a petition that the petitioner reasonably could have raised); Cobalt Boats, LLC v. Sea Ray Boats, Inc., Case No. 15-cv-21, 2017 WL 2605977, at *3 (E.D. Va. June 5, 2017) (same); Biscotti Inc. v. Microsoft Corp., Case No. 13-cv-1015, 2017 WL 2526231, at *7 (E.D. Tex. May 11, 2017) (same); Douglas Dynamics, LLC v. Meyer Prods. LLC, Case No. 14-cv-886, 2017 WL 1382556, at *5 (W.D. Wis. Apr. 18, 2017) (same); Parallel Networks Licensing, LLC v. IBM Corp., Case No. 13-cv- 2072, 2017 WL 1045912, at *12 (D. Del. Feb. 22, 2017) (same); Oil-Dri Corp. of Am. v. Nestle Purina Petcare Co., Case No. 15-cv-1067, 2017 WL 3278915, at *8 (N.D. Ill. Aug. 2, 2017) (“[W]hile it makes sense that noninstituted grounds do not give rise to estoppel because a petitioner cannot—to no fault of its own—raise those grounds after the institution decision, when a petitioner simply does not Case: 20-2222 Document: 63 Page: 22 Filed: 02/04/2022

raise invalidity grounds it reasonably could have raised in an IPR petition, the situation is different.”). Other district courts read Shaw differently, focusing on Shaw’s discussion of the “during the IPR” language in § 315(e)(2). See, e.g., Koninklijke Philips N.V. v. Wangs All. Corp., Case No. 14-cv-12298, 2018 WL 283893, at *4 (D. Mass. Jan. 2, 2018) (“It would seem, then, that the phrase “inter partes review” . . . refers only to the period of time after review is instituted, and, therefore, the estoppel pro- vision does not apply to arguments that the petitioner only ‘raised or reasonably could have raised’ in its petition ra- ther than after institution of review.”); Verinata Health, Inc. v. Ariosa Diagnostics, Inc., Case No. 12-cv-5501, 2017 WL 235048, at *3 (N.D. Cal. Jan 19, 2017) (“The [Shaw] court chose instead to interpret the IPR estoppel language literally, plainly stating that only arguments raised or that reasonably could have been raised during IPR are subject to estoppel.”); Intellectual Ventures I LLC v. Toshiba Corp., 221 F. Supp. 3d 534, 553–54 (D. Del. 2016) (holding that although exempting nonpetitioned grounds from estoppel “confounds the very purpose of this parallel administrative proceeding, the court cannot divine a reasoned way around the Federal Circuit’s interpretation in Shaw”). After Shaw, in SAS Institute, Inc. v. Iancu, 138 S. Ct. 1348 (2018), the Supreme Court made clear both that there is no partial institution authority conferred on the Board by the America Invents Act and that it is the petition, not the institution decision, that defines the scope of the IPR. See id. at 1357–58 (“[T]he statute tells us that the peti- tioner’s contentions, not the Director’s discretion, define the scope of the litigation . . . There is no room in this scheme for a wholly unmentioned ‘partial institution’ power that lets the Director select only some challenged claims for decision.”). Given the statutory interpretation in SAS, any ground that could have been raised in a peti- tion is a ground that could have been reasonably raised Case: 20-2222 Document: 63 Page: 23 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 23

“during inter partes review.” Thus, the Supreme Court’s later decision in SAS makes clear that Shaw, while per- haps correct at the time in light of our pre-SAS interpreta- tion of the statute cannot be sustained under the Supreme Court’s interpretation of related statutory provisions in SAS. The panel here has the authority to overrule Shaw in light of SAS, without en banc action. To be sure, SAS did not explicitly overrule Shaw or address the scope of statu- tory estoppel under § 315(e)(2). But the reasoning of Shaw rests on the assumption that the Board need not institute on all grounds, an assumption that SAS rejected. Even in the Ninth Circuit, which has one of the stricter approaches to panel overruling, see Henry J. Dickman, Conflicts of Precedent, 106 Va. L. Rev. 1345, 1350–51 (2020), “the is- sues decided by the higher court need not be identical in order to be controlling. Rather, the relevant court of last resort must have undercut the theory or reasoning under- lying the prior circuit precedent in such a way that the cases are clearly irreconcilable,” Miller v. Gammie, 335 F.3d 889, 900 (9th Cir. 2003) (en banc). We approved that higher standard in Troy v. Samson Manufacturing Corp., 758 F.3d 1322, 1326 (Fed. Cir. 2014), and conclude that that standard is satisfied in this case. Accordingly, we take this opportunity to overrule Shaw and clarify that estoppel applies not just to claims and grounds asserted in the petition and instituted for consid- eration by the Board, but to all claims and grounds not in the IPR but which reasonably could have been included in the petition. 5 In a regime in which the Board must

5 In this case, SAS was decided while IPR proceed- ings remained pending before the Board. Accordingly, we need not decide the scope of preclusion in cases in which Case: 20-2222 Document: 63 Page: 24 Filed: 02/04/2022

institute on all grounds asserted and the petition defines the IPR litigation, this interpretation is the only plausible reading of “reasonably could have been raised” and “in the IPR” that gives any meaning to those words. It is undisputed that Apple and Broadcom were aware of the prior art references that they sought to raise in the district court when Apple filed its IPR petitions. Despite not being included in any of Apple’s IPR petitions, the con- tested grounds reasonably could have been included in the petitions, and thus in the IPR. We affirm the district court’s decision barring Apple and Broadcom from raising invalidity challenges based on these prior art references. IV. Inequitable Conduct We turn next to the district court’s grant of Caltech’s summary judgment motion of no inequitable conduct. Generally, inequitable conduct requires a showing that un- disclosed prior art was but-for material to the PTO’s deci- sion of patentability. Therasense, Inc. v. Becton, Dickinson and Co., 649 F.3d 1276, 1291 (Fed. Cir. 2011). Prior art is but-for material if the PTO would have denied a claim had it known of the undisclosed prior art. Id. Prior art is not but-for material if it is merely cumulative. Regeneron Pharms., Inc. v. Merus N.V., 864 F.3d 1343, 1350 (Fed. Cir. 2017) (citing Dig. Control Inc. v. Charles Mach. Works, 437 F.3d 1309, 1319 (Fed. Cir. 2006)). Broadcom and Apple on appeal have limited their ar- gument to the district court’s conclusion that Richardson99 was not shown to be but-for material to patentability. The district court found that Richardson99 was merely cumu- lative of Luby97 and Luby98—references the PTAB consid- ered in IPR proceedings upholding the patents’ validity—

the Board declined to institute on all grounds and issued its final written decision pre-SAS. Case: 20-2222 Document: 63 Page: 25 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 25

noting Apple and Broadcom’s pleadings, interrogatory re- sponses, and briefs failed to distinguish Luby’s disclosed ir- regular repetition from Richardson99’s. Apple and Broadcom did not argue at the summary judgment stage that Richardson99 was different from Luby such that it was not merely cumulative. The district court rejected the arguments as to Richardson99 because the Appellants failed to put Caltech on notice of an independent inequita- ble conduct theory based on alleged differences between Richardson99 and Luby. The district court’s decision was not an abuse of discre- tion. We therefore affirm the grant of summary judgment of no inequitable conduct. V. Damages Caltech presented to the jury a two-tier reasonable roy- alty model based on simultaneous hypothetical negotia- tions with Broadcom and Apple in December 2009. Broadcom and Apple generally argue that the damages judgement cannot be sustained because Caltech’s damages model impermissibly applied two separate hypothetical ne- gotiations for Broadcom and Apple for sales of the same chips; because Caltech’s royalty rates were derived from non-comparable settlements—without apportionment and based on improperly excluded expert opinions and unre- lated “black box” calculations; and because Caltech’s dam- age model improperly included extraterritorial sales. Caltech argues in response that the damages judgment properly rests on separate running-royalty rates for each defendant, that the district court’s rulings on admissibility and exclusion of evidence were not an abuse of discretion and that the damages were based entirely on United States sales. We find no error in the district court’s jury instructions relating to extraterritoriality. But because Caltech’s two- tier damages theory is legally unsupportable on this Case: 20-2222 Document: 63 Page: 26 Filed: 02/04/2022

record, the damages award is vacated and the case is re- manded for a new trial on damages. A. Extraterritoriality Apple and Broadcom argue that the damages verdict improperly included extraterritorial sales from Broadcom’s international affiliates. They argue that the district court erroneously instructed the jury on extraterritoriality for two reasons. First, they argue that the district court erro- neously declined to instruct the jury of a presumption against extraterritorial application of United States laws. We see no error. The relevant presumption is whether a law applies extraterritorially. See WesternGeco LLC v. ION Geophysical Corp., 138 S.Ct. 2129, 2134, 2136 (2018). But the dispute here is not whether infringement laws ap- ply domestically or extraterritorially—there is no dispute that the laws apply only domestically. Rather, the dispute between the parties is whether the relevant transactions here were domestic or extraterritorial in nature. The pre- sumption against extraterritorial application is thus inap- plicable. As Caltech correctly argues, the district court expressly instructed the jury that Caltech had the burden of proving that infringement occurred in the United States. J. App’x 184–85 (instructing the jury that “An allaged in- fringer is liable for direct infringement of a claim if the pa- tent holder proves by a preponderance of the evidence that the infringer, without the patent holder’s authorization, imports, offers to sell, sells, or uses [the accused products] within the United States,” and explaining the factors for de- termining whether a sale occurs in the United States). This was a proper and sufficient jury instruction with re- spect to the applicable burdens on the territoriality of the sales at issue. Second, Apple and Broadcom argue that the district court erroneously instructed the jury that the “sales cycle leading to design wins” could trigger a United States sale. Apple and Broadcom argue that Halo recognized a Case: 20-2222 Document: 63 Page: 27 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 27

categorical prohibition against treating such a sales cycle as a domestic sale. See Halo Elecs., Inc. v. Pulse Elecs., Inc., 831 F.3d 1369, 1378 (Fed. Cir. 2016), on remand from 579 U.S. 93 (2016). Halo held that “pricing and contracting negotiations in the United States alone do not constitute or transform those extraterritorial activities into a sale within the United States for purposes of § 271(a).” Id. It held that this was so “when substantial activities of a sales transaction, including the final formation of a contract for sales encompassing all essential terms as well as the deliv- ery and performance under that contract, occur entirely outside the United States.” Id. This is not a blanket hold- ing that design wins arising out of a sales cycle can never be domestic transactions. Indeed, the district court noted that a design win meeting these criteria, such that “sub- stantial activities of a sales transaction . . . occurs entirely outside the United States” would not constitute a sale within the United States. J. App’x 185. The district court’s jury instruction emphasized the key question of whether there were such substantial activities in the United States, an instruction that Apple and Broadcom do not contest. See also Carnegie Mellon U. v. Marvell Tech. Grp., Ltd., 807 F.3d 1283, 1308 (Fed. Cir. 2015). There is no error in the district court’s permissive instruction. B. Caltech’s Two-Tier Damage Model Caltech presented its damage theory to the jury through two experts, Dr. Catherine Lawton and Dr. David Teece. They opined that Caltech would have engaged in two simultaneous hypothetical negotiations, one with Broadcom at the “chip level” and one with Apple at the “de- vice level.” Those negotiations would have excluded from Broadcom’s hypothetical chip license any Broadcom chips incorporated into Apple products sold in the United States and treated those identical chips as being subject to Apple’s separate hypothetical device license at a vastly different royalty rate. Both of Caltech’s experts testified that sepa- rate chip-level and device-level negotiations would have Case: 20-2222 Document: 63 Page: 28 Filed: 02/04/2022

been proper, rather than a single hypothetical negotiation for all of the accused chips, because both defendants were separate infringers and there would be no “cross-talk” be- tween them as they each engaged in their own hypothetical negotiation. The district court considered the opinions of Caltech’s experts and, over Broadcom and Apple’s objection, permit- ted Caltech to present that theory to the jury. In doing so, the district court observed that “[p]atent owners will some- times seek damages from accused infringers at different levels in the supply chain, and so long as they do not at- tempt to obtain a double recovery to violate other legal principles like patent exhaustion, they are free to do so.” J. App’x 225. In ruling in Caltech’s favor, the district court saw no concern over double recovery because Broadcom and Apple were different companies and because the ex- perts’ opinions carved out of the Broadcom hypothetical ne- gotiation chips sold to Apple. But in the absence of some evidence that companies in the positions of Broadcom and Apple would engage in such separate negotiations and in the absence of additional facts that might justify separate and different treatment of the same chips at different lev- els of the supply chain, the mere fact that Broadcom and Apple are separate infringers alone does not support treat- ing the same chips differently at different stages in the sup- ply chain and does not justify submitting such a two-tier damage theory to the jury. It is generally recognized that in the usual case, “a direct infringer or someone who in- duced infringement should pay the same reasonable roy- alty based on a single hypothetical negotiation analysis.” LaserDynamics, Inc. v. Quanta Comput., Inc., 694 F.3d 51, 76 (Fed. Cir. 2012). Caltech argued that separate royalty rates at different levels of the supply chain are proper because the reasona- ble royalty inquiry focuses on the amount of value that the patent technology adds to a product, citing Ericsson, Inc. v. Case: 20-2222 Document: 63 Page: 29 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 29

D-Link Sys., 773 F.3d 1201, 1226 (Fed. Cir. 2014). The dis- trict court concluded that Broadcom and Apple’s products were different and therefore possessed different values simply because Broadcom and Apple were “different com- panies at different levels in the supply chain.” J. App’x 226. But to reach that conclusion without more ignores estab- lished precedent to the effect that, in the absence of a com- pelling showing otherwise, a higher royalty is not available for the same device at a different point in the supply chain. As we previously held, “a reasonable royalty is not to be separately calculated against each successive infringer. Once full recovery is obtained from one infringer with re- spect to a particular infringing device, at most nominal ad- ditional damages may be awarded against another with respect to the same device.” Stickle v. Heublein, Inc., 716 F.2d 1550, 1562 (Fed. Cir. 1983). Moreover, “[a] party is precluded from suing to collect damages from direct in- fringement by a buyer and user of a product when actual damages covering that very use have already been col- lected from the maker and seller of that product.” Glenayre Elecs., Inc. v. Jackson, 443 F.3d 851, 864 (Fed. Cir. 2006). The district court cited but distinguished those cases as only applying to damages calculations against two de- fendants involving overlapping royalty bases, a situation not existing here based on Caltech’s expert’s exclusion of chips sold to Apple from the royalty base considered for Broadcom. But that exclusion in this case is wholly con- trived, lacks any basis of fact and is contrary to the custom- ary way patent infringement disputes are ordinarily resolved. It is well settled that a reasonable royalty is what a willing licensor and a willing licensee would have agreed to at a hypothetical negotiation just before infringement began. See Carnegie Mellon Univ. v. Marvell Tech. Grp., 807 F.3d 1283, 1303-1304 (Fed. Cir. 2015). Here, there is nothing in the record to suggest that Broadcom and Apple would have been willing to negotiate in this artificial way rather than to more conventionally negotiate a single Case: 20-2222 Document: 63 Page: 30 Filed: 02/04/2022

license at a single rate for the same chips. Neither of Cal- tech’s experts offered any factual basis to conclude that Broadcom and Apple would have been willing to engage in separate negotiations leading to vastly different royalty rates for the same chips. The district court’s views to the contrary and its limiting of the Stickle and Glenayre cases to situations involving double recovery were misplaced and erroneous. Caltech’s two-tier damages theory is legally un- supportable on this record. *** We need not and do not address Broadcom and Apple’s indemnification argument, or their argument that the hy- pothetical negotiations would have been held not with Cal- tech but by its exclusive licensee, Inforon. Nor do we address Broadcom and Apple’s argument based on small- est-saleable-patent-practicing-unit, or the sufficiency of the evidence as to the domestic or extraterritorial charac- ter of Broadcom’s sales.

CONCLUSION For the foregoing reasons, we affirm the district court’s construction of the claim limitation “repeat.” We affirm the district court’s denial of JMOL on infringement of the as- serted claims of the ’710 and ’032 patents. We affirm the district court’s conclusion that claim 13 of the ’781 patent is patent-eligible but vacate the jury’s verdict of infringe- ment thereof and remand for a new trial. We affirm the district court’s summary judgment findings of no invalidity based on IPR estoppel and no inequitable conduct. We af- firm the district court’s jury instructions relating to extra- territoriality, but vacate the jury’s damage award and remand for a new trial on damages. AFFIRMED IN PART, VACATED IN PART, AND REMANDED. Case: 20-2222 Document: 63 Page: 31 Filed: 02/04/2022

THE CALIFORNIA INSTITUTE v. BROADCOM LIMITED 31

COSTS Each party shall bear its own costs. Case: 20-2222 Document: 63 Page: 32 Filed: 02/04/2022

United States Court of Appeals for the Federal Circuit ______________________

BROADCOM LIMITED, NKA BROADCOM INC., BROADCOM CORPORATION, AVAGO TECHNOLOGIES LIMITED, NKA AVAGO TECHNOLOGIES INTERNATIONAL SALES PTE. LIMITED, APPLE INC., Defendants-Appellants ______________________

Appeals from the United States District Court for the Central District of California in No. 2:16-cv-03714-GW- AGR, Judge George H. Wu. ______________________

DYK, Circuit Judge, concurring-in-part and dissenting-in- part. While I join Discussion sections II.B.1, III, and IV of the majority opinion, I respectfully disagree with the ma- jority’s holding that substantial evidence supports the jury’s verdict of infringement of the asserted claims of the ’710 and ’032 patents and would reverse the district court’s denial of JMOL of no literal infringement. I would simi- larly reverse the denial of JMOL rather than remand for a Case: 20-2222 Document: 63 Page: 33 Filed: 02/04/2022

new trial with respect to the infringement of the ’781 pa- tent. I No matter how novel an invention is, it is the patent’s claims that “determine what the invention is,” as well as the bounds of the patent owner’s rights to that invention’s “exclusive use.” Motion Picture Patents Co. v. Universal Film Mfg. Co., 243 U.S. 502, 510 (1917). Patent owners are limited by the language in the claims, and “can claim noth- ing beyond them.” Id. (quoting Keystone Bridge Co. v. Phoenix Iron Co., 95 U.S. 274, 278 (1877)). It is the patent owner’s burden to show that “the properly construed claim reads on the accused device exactly.” CommScope Techs. LLC v. Dali Wireless Inc., 10 F.4th 1289, 1295 (Fed. Cir. 2021) (quoting Engel Indus., Inc. v. Lockformer Co., 96 F.3d 1398, 1405 (Fed. Cir. 1996)). Denial of JMOL of no literal infringement must be reversed when plaintiffs “fail[] to present evidence proving that the [accused device] meets the district court’s construction of [a] claim term.” CommScope, 10 F.4th at 1295. Such a failure occurs when expert testimony “points to the result . . . rather than the specific mechanism claimed to achieve that result,” id. at 1297, or when expert testimony is merely “cursory,” Alexsam, Inc. v. IDT Corp., 715 F.3d 1336, 1341–42 (Fed. Cir. 2013). Here, no showing of literal infringement has been made that can support the jury’s infringement verdict under the district court’s claim construction, which is not challenged by Caltech. II Before trial, the district court held a Markman hearing to construe the term “repeat,” an essential limitation in all of the asserted claims of the three patents. Caltech advo- cated for the term’s plain and ordinary meaning. Apple and Broadcom proposed a narrower construction, contend- ing that “repeat” should be construed as “creating a new bit that corresponds to the value of an original bit (i.e., a new Case: 20-2222 Document: 63 Page: 34 Filed: 02/04/2022

copy) by storing the new copied bit in memory. A reuse of a bit is not a repeat of a bit.” J.A. 9. The district court rejected this construction because reusing bits by “selecting the bits for use without necessarily storing them at a spe- cific location in computer memory” could satisfy the claim limitation. J.A. 10. The district court nonetheless noted that the “claim language . . . makes clear that ‘repeated bits’ are a construct distinct from the original bits from which they are created.” Id. Ultimately, the district court adopted the term’s plain and ordinary meaning, concluding that no further construction was required. Following the Markman hearing, the parties continued to dispute what exactly the plain and ordinary meaning of “repeat” entailed. As a result, at trial, as permitted by precedent, 1 the district court revisited and clarified its ear- lier claim construction ruling, instructing the jury that the ’710 and ’032 patents’ claimed repetition requires the “gen- eration of additional bits, where generation can include, for example, duplication or reuse of bits.” J.A. 171. The criti- cal question, therefore, is whether there is substantial evi- dence that the accused devices cause “generation of additional bits.” Unfortunately, in denying the appellants’ post-trial JMOL motion, the district court provided no analysis of how Caltech established infringement, relegat- ing this question to a footnote which said only that

1 It is within the district court’s discretion to “engage in a rolling claim construction, in which the court revisits and alters its interpretation of the claim terms as its un- derstanding of the technology evolves.” See, e.g., Jack Guttman, Inc. v. Kopykake Enters., 302 F.3d 1352, 1361 (Fed. Cir. 2002) (citing Sofamor Danek Grp., Inc. v. DePuy– Motech, Inc., 74 F.3d 1216, 1221 (Fed. Cir. 1996)); Col- legeNet, Inc v. ApplyYourself, Inc., 418 F.3d 1225, 1233–34 (Fed. Cir. 2005). Case: 20-2222 Document: 63 Page: 35 Filed: 02/04/2022

“Defendants’ arguments that the verdict was not supported by substantial evidence also remain unpersuasive.” J.A. 206. The panel majority does not identify or rely on a reuse theory to uphold the jury’s verdict. Rather, the majority concludes that infringement of the repeat limitation was supported by expert testimony that “the [AND] gate affirm- atively enables the information bit to be duplicated as the output bit. That is a ‘repeat’ . . . because the information bit in that situation ‘flows through’ to appear again in the output.” Maj. Op. at 16 (emphasis added). But there is in fact no such expert testimony. To the contrary—consistent with its claim construction position, Caltech’s expert testi- fied that the claims do not “require that the repeat has to be done by duplicating information bits,” J.A. 2858, and Caltech argued to this court on appeal that “repetition does not require duplication,” Appellee’s Br. 18. To be sure, Caltech is correct that duplication is not re- quired to satisfy the repeat limitation. But the problem for Caltech (and for the majority) is that Caltech never estab- lished that the accused devices generate “additional bits,” as required by the district court’s claim construction. The infringement theory presented at trial explained that the accused devices work as follows: information bits are input into the accused devices, those bits travel down branched wires to the inputs of 972 AND gates, and three to twelve of those AND gates will be open for each information bit, thus outputting the bits a different number of times. For this theory to satisfy Caltech’s burden, Caltech was re- quired to establish where, when, and how additional bits were generated. One possibility—presented by Caltech’s counsel—was that additional bits were generated by branching at the in- puts. During closing arguments, Caltech told the jury “That’s how you repeat bits, with a voltage along wires . . . how else would you repeat bits? How else would you do it?” Case: 20-2222 Document: 63 Page: 36 Filed: 02/04/2022

1/28 Tr. 90:20–91:9. But the record does not support a the- ory that the branched wires generate additional bits. Cal- tech’s experts testified merely that the bits are “connected to” the AND gates by branched wires, without explaining whether or how that connection generated additional bits. J.A. 2831. Apple and Broadcom presented unrefuted ex- pert testimony that the branched wire connection involves simultaneously sending the same bit—not an additional bit—to the inputs of AND gates. Caltech’s expert did not testify to the contrary, and in fact declined to testify that branching generates additional bits. Q. Branch wire creates repeat bits? ... THE COURT: I’ll allow him to answer the question if he understands it. THE WITNESS: That question I did not under- stand. I didn’t think it was well formed. BY MR. MUELLER: Q. The branch wire in the Broadcom chips in your view creates repeat bits within the meaning of the claims; correct? ... THE WITNESS: In my analysis the branch wire is being used in conjunction with the tables and these and gates to implement irregular repetition. J.A. 3019–20. There is no substantial evidence supporting an infringement verdict based on branching. Caltech’s separate theory to establish that the accused devices generate additional bits was the “flow through” theory, supported, according to the majority, by expert tes- timony that “the information bit ‘flows through’ to the out- put gate [when] the parity-check bit is 1,” and that the flow through bit constitutes “a repeat, both according to the ex- pert’s usage and a plain understanding of the word.” Maj. Case: 20-2222 Document: 63 Page: 37 Filed: 02/04/2022

Op. at 16–17. The majority identifies record evidence where Caltech’s expert summarily testified that the AND gates repeated information bits when the gates are ena- bled, J.A. 2842; J.A. 3080, and where the expert concluded that the AND gates were “generating additional bits at their output,” J.A. 4162. This cursory and conclusory tes- timony cannot satisfy Caltech’s burden. Caltech’s own expert testimony as to how an AND gate functions forecloses concluding that the AND gates gener- ate additional bits. Throughout trial, Caltech’s expert con- sistently testified that an AND gate “act[s] like a switch . . . allowing the information bits to flow through . . . or not.” J.A. 3016–17; see also J.A. 3031 (“[E]nable one allows the AND gate to act like a switch. So enable would close the switch and allow the one to come out.”); J.A. 3030 (“[T]he AND gate has two inputs. One of them is the enable that allows the switch to open or close.”); J.A. 3031 (“Enable zero is like the switch not allowing the information bit to flow through.”). If an enabled AND gate merely allows the same information bit that already exists at the input of an AND gate “to come out,” J.A. 3031, or to “flow through,” this does not remotely establish how the AND gate output generates additional information bits. Under the explanation that Caltech repeatedly presented, one information bit comes in and one information bit comes out—no additional bit ap- pears anywhere. Flow through, with a 1:1 ratio of input bits to output bits as described by Caltech’s experts, cannot satisfy the generation of additional bits limitation required by the district court’s claim construction. Caltech’s remaining theory submits that the accused devices generate additional bits when the branched wires are combined with the AND gates. The majority believes that “the physical connection of the first inputs of all 972 AND gates . . . and the connection of the parity-bit system to the other inputs of the AND gates to selectively enable 3 to 12 of those gates . . . together implement irregular repe- tition.” Maj. Op. at 17. Neither the majority’s opinion, Case: 20-2222 Document: 63 Page: 38 Filed: 02/04/2022

Caltech’s briefs, and most importantly, Caltech’s experts, explain why the combination of these two non-infringing components results in infringement. The district court therefore erred in denying JMOL of no literal infringe- ment. III At trial, Caltech also presented a doctrine of equiva- lents case to the jury. In a footnote to its JMOL decision, the district court found that it was “not necessary” to ana- lyze Apple and Broadcom’s challenge to the doctrine of equivalents arguments, J.A. 206, and the majority here similarly does not address such a theory. There is no basis to sustain the verdict on a doctrine of equivalents theory. To prevail, Caltech had the burden of proving equiva- lence “between the elements of the accused product or pro- cess and the claimed elements of the patented invention,” Warner-Jenkinson Co. v. Hilton Davis Chem. Co., 520 U.S. 17, 21 (1997), by showing that “the accused device contains an element that is not ‘substantially different’ from any claim element that is literally lacking,” Kraft Foods, Inc. v. Int’l Trading Co., 203 F.3d 1362, 1371 (Fed. Cir. 2000) (cit- ing Warner-Jenkinson, 520 U.S. at 40). Caltech’s appellate brief devoted only one and a half pages to this issue, citing to eight pages of trial testimony in which its expert asserted that the differences between the accused chips and the claim limitations were insub- stantial. At trial, Dr. Shoemake testified that because the claims do not “require repeating with any type of specific circuitry,” any differences in the method Broadcom’s chips used to accomplish the “overall goal” of the claims were in- substantial. J.A. 2856–58. But even if no specific circuitry is required, Dr. Shoe- make never explained why a bit flowing through to the out- put of an AND gate is substantially similar to the claimed device that generates additional bits. This is reminiscent Case: 20-2222 Document: 63 Page: 39 Filed: 02/04/2022

of the insufficient, “[g]eneralized testimony” proscribed by Texas Instruments Inc. v. Cypress Semiconductor Corp., 90 F.3d 1558, 1567 (Fed. Cir. 1996), and it cannot satisfy Cal- tech’s burden to provide “particularized testimony and linking argument as to the ‘insubstantiality of the differ- ences’ between the claimed invention and the accused de- vice or process . . . on a limitation-by-limitation basis,” id. There was no basis for the jury to find infringement under the doctrine of equivalents. I would reverse the district court’s denial of JMOL and enter judgment of non-infringement for Broadcom and Ap- ple.