Intel Corporation v. Qualcomm Incorporated

Case: 20-1828 Document: 98 Page: 1 Filed: 12/28/2021 United States Court of Appeals for the Federal Circuit ______________________ INTEL CORPORATION, Appellant v. QUALCOMM INCORPORATED, Cross-Appellant ______________________ 2020-1828, 2020-1867 ______________________ Appeals from the United States Patent and Trademark Office, Patent Trial and Appeal Board in Nos. IPR2018- 01334, IPR2018-01335, IPR2018-01336. ______________________ Decided: December 28, 2021 ______________________ THOMAS SAUNDERS, Wilmer Cutler Pickering Hale and Dorr LLP, Washington, DC, argued for appellant. Also rep- resented by DAVID LANGDON CAVANAUGH, CLAIRE HYUNGYO CHUNG; JOSEPH F. HAAG, Palo Alto, CA. JENNIFER L. SWIZE, Jones Day, Washington, DC, ar- gued for cross-appellant. Also represented by ROBERT BREETZ, DAVID B. COCHRAN, DAVID MICHAEL MAIORANA, JOSEPH M. SAUER, Cleveland, OH; KELLY HOLT, New York, NY; MATTHEW JOHNSON, JOSHUA R. NIGHTINGALE, Pitts- burgh, PA; ISRAEL SASHA MAYERGOYZ, Chicago, IL. ______________________ Case: 20-1828 Document: 98 Page: 2 Filed: 12/28/2021 2 INTEL CORPORATION v. QUALCOMM INCORPORATED Before PROST, TARANTO, and HUGHES, Circuit Judges. TARANTO, Circuit Judge. Qualcomm Inc. owns U.S. Patent No. 8,838,949, which addresses multi-processor systems in which software stored in non-volatile memory coupled to a first processor is to be used by a second processor. The patent describes and claims systems, methods, and apparatuses for effi- ciently retrieving an executable software image from the first processor’s non-volatile memory and loading it for use by the second processor. Intel Corp. challenged all claims of the ’949 patent as unpatentable for obviousness in three inter partes reviews (IPRs) before the Patent and Trade- mark Office. The Office’s Patent Trial and Appeal Board consolidated the proceedings and issued a final written de- cision holding that Intel had proved unpatentable claims 10, 11, 13–15, and 18–23, but not claims 1–9, 12, 16, and 17. Intel Corp. v. Qualcomm Inc., IPR2018-01334, 2020 WL 1286306, at *27 (P.T.A.B. Mar. 16, 2020) (Final Writ- ten Decision). Intel appeals. We hold first that Intel has adequately demonstrated Article III standing to press this appeal. On the merits, we hold that in the decision before us, the Board failed to tie its construction of the phrase “hardware buffer” to the ac- tual invention described in the specification. For that rea- son, we vacate the Board’s decision as to claims 1–9 and 12 and remand for a new construction. As to claims 16 and 17, which are in means-plus-function format, we also va- cate and remand. We conclude that the Board failed to de- termine for itself whether there is sufficient corresponding structure in the specification to support those claims and whether it can resolve the patentability challenges despite the (potential) indefiniteness of those claims. Case: 20-1828 Document: 98 Page: 3 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 3 I A The patent addresses a system with multiple proces- sors, each of which must execute its own “boot code” to play its operational role in the system. Such code must be stored in non-volatile memory (e.g., flash memory or read- only memory), since volatile memory is cleared when the device powers down; and the boot code generally must be transferred to its corresponding processor’s volatile memory in order to be executed by that processor. ’949 pa- tent, col. 1, lines 39–41. In a multi-processor system, one possible design choice is to store the boot code for each pro- cessor in its own separate non-volatile memory. Another choice, to avoid the costs of multiple memories each ade- quate for such storage, is to store the boot code for one pro- cessor in the non-volatile memory of another processor, permitting elimination or shrinkage of the non-volatile memory of the first processor. Id., col. 1, line 60, through col. 2, line 14. The ’949 patent, titled “Direct Scatter Loading of Exe- cutable Software Image from a Primary Processor to One or More Secondary Processor in a Multi-Processor System,” assumes the latter, shared-storage choice. It addresses the problem, inherent in that choice, of loading the boot code for a “secondary” processor (into its volatile memory) from the non-volatile memory of a “primary” processor. Id., col. 2, line 58, through col. 3, line 2. It uses a “direct scatter load” procedure to do so. “Scatter loading” refers to moving a “binary multi-segmented” software image into scattered parts (as opposed to one contiguous block) of the secondary processor’s “system memory” before executing it. Id., col. 2, lines 14–22. The patent discloses a “direct” scatter load- ing process, through which the segments of the software image are transmitted “directly” from a “hardware buffer” to their final locations in the secondary processor’s “system memory.” Id., col. 2, lines 58–63. Case: 20-1828 Document: 98 Page: 4 Filed: 12/28/2021 4 INTEL CORPORATION v. QUALCOMM INCORPORATED Claims 1 and 2 are representative for the claim-con- struction issue on appeal. They recite: 1. A multi-processor system comprising: a secondary processor comprising: system memory and a hardware buffer for receiving an image header and at least one data seg- ment of an executable software im- age, the image header and each data segment being received sepa- rately, and a scatter loader controller configured: to load the image header, and to scatter load each received data segment based at least in part on the loaded image header, di- rectly from the hardware buffer to the system memory; a primary processor coupled with a memory, the memory storing the executa- ble software image for the secondary pro- cessor; and an interface communicatively coupling the primary processor and the secondary pro- cessor, the executable software image be- ing received by the secondary processor via the interface. 1 1 The indentation of the last two components of the “multi-processor system” (the “primary processor” and the “interface”) has been altered from the original to reflect the Case: 20-1828 Document: 98 Page: 5 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 5 2. The multi-processor system of claim 1 in which the scatter loader controller is configured to load the executable software image directly from the hardware buffer to the system memory of the secondary processor without copying data be- tween system memory locations on the second- ary processor. Id., col. 12, line 60, through col. 13, line 16 (emphases added). Claim 16 is relevant to the means-plus-function issue on appeal. It recites: 16. An apparatus comprising: means for receiving at a secondary processor, from a primary processor via an inter-chip communica- tion bus, an image header for an executable soft- ware image for the secondary processor that is stored in memory coupled to the primary processor, the executable software image comprising the im- age header and at least one data segment, the im- age header and each data segment being received separately; means for processing, by the secondary proces- sor, the image header to determine at least one lo- cation within system memory to which the secondary processor is coupled to store each data segment; means for receiving at the secondary processor, from the primary processor via the inter-chip com- munication bus, each data segment; and fact that they are parts of the multi-processor system, not parts of the secondary processor. Case: 20-1828 Document: 98 Page: 6 Filed: 12/28/2021 6 INTEL CORPORATION v. QUALCOMM INCORPORATED means for scatter loading, by the secondary pro- cessor, each data segment directly to the deter- mined at least one location within the system memory, and each data segment being scatter loaded based at least in part on the processed im- age header. Id., col. 14, lines 17–37 (emphases added). B In 2017, Qualcomm initiated actions against Apple Inc. (not party to this appeal) in district court and at the Inter- national Trade Commission (ITC), alleging that Apple in- fringed the ’949 patent (and other patents) by making, selling, and using iPhone models that incorporated base- band processors made by Intel. See J.A. 6233. Qualcomm’s infringement theory identified the Intel baseband proces- sors as the “secondary processors” of the ’949 patent and pointed to “Intel design documents that describe how the Intel baseband processors are integrated into the iPhones.” See J.A. 6262, 6264. In 2019, Qualcomm and Apple settled all litigation worldwide between the two companies, and Qualcomm agreed to license the patents to Apple for six years (and two additional years if Apple wished). J.A. 6084, 6210. Later in 2019, Apple acquired “the majority of Intel’s smartphone modem business” (including its produc- tion of baseband processors), J.A. 6088, though Intel con- tinues to supply pre-acquisition versions of its baseband processors to Apple and another customer, J.A. 6204. Meanwhile, in 2018, Intel petitioned the Board for three inter partes reviews, each petition covering various claims of the ’949 patent and together covering all. After consolidating the reviews, the Board issued a final written decision. The Board ruled that Intel had proved the un- patentability of claims 10, 11, 13–15, and 18–23, but Qual- comm, despite filing a cross-appeal to raise the issue, no longer challenges that ruling. The Board also ruled that Intel had failed to prove the unpatentability of the Case: 20-1828 Document: 98 Page: 7 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 7 remaining claims: (a) independent claim 1 and its depend- ent claims 2–9, plus claim 12 (depending on independent claim 10); and (b) independent claim 16 and its dependent claim 17. See Final Written Decision, at *27. Intel appeals those losses. The Board’s determination upholding claims 1–9 and 12 turned on the construction of the claim phrase “hard- ware buffer.” No party put forward an explicit construction of that term until Qualcomm’s patent owner response, in which Qualcomm proposed to construe the phrase to mean “a buffer within a hardware transport mechanism that re- ceives data sent from the primary processor to the second- ary processor.” J.A. 4224. In reply, Intel argued that “‘hardware buffer’ should be given its ordinary meaning of ‘a buffer implemented in hardware.’” J.A. 4322. In sur- reply, Qualcomm defended its proposed construction but also advanced an alternative construction, “a buffer that is not allocated by the secondary processor.” J.A. 4397. Qual- comm explained: “In the ’949 patent, the hardware buffer is a permanent buffer within the hardware transport mech- anism, in contrast to a temporary buffer in system memory that is allocated by the secondary processor at run time for this purpose.” Id. (cleaned up). The Board rejected both Qualcomm’s original construc- tion, which referred to “a hardware transport mechanism,” and Intel’s “ordinary meaning” construction, “a buffer im- plemented in hardware.” Final Written Decision, at *5–6. Instead, the Board agreed with Qualcomm that “the ’949 patent does differentiate disclosed loading techniques from known prior art techniques that use temporary buffers” and concluded that “the ‘hardware buffer’ limitations . . . ‘should not be read so broadly as to encompass’ the use of a temporary buffer.” Id. at *7 (quoting SciMed Life Systems, Inc. v. Advanced Cardiovascular Systems, Inc., 242 F.3d 1337, 1343 (Fed. Cir. 2001)). Case: 20-1828 Document: 98 Page: 8 Filed: 12/28/2021 8 INTEL CORPORATION v. QUALCOMM INCORPORATED That construction, the Board concluded, was disposi- tive of the prior-art challenges to claims 1–9 and 12. Intel had argued that the required “hardware buffer” was taught by the “intermediate storage area” disclosed in its principal prior-art reference (Svensson, U.S. Patent No. 7,356,680). But because Svensson’s “intermediate storage area” was “reserved at runtime,” the Board determined, it was a tem- porary buffer and thus not a “hardware buffer.” Id. at *23– 24. As to claims 16 and 17, the Board ruled that Intel had failed to show unpatentability because Intel had not met its burden of identifying sufficient structure corresponding to two means-plus-function terms in independent claim 16. Claim 16 recites a “means for processing” and a “means for scatter loading.” Id. at *7–8. In IPR2018-01335, Intel’s petition identified those phrases as means-plus-function limitations and offered the same constructions for them that Qualcomm had offered in prior ITC litigation. In its institution decision, the Board wrote that it “ha[d] ques- tions as to the sufficiency of [Intel’s] identified structures” for the two terms. J.A. 5160. But because the petition met the standard for institution on claims 10–15, the Board in- stituted review on claims 16 and 17 as well, suggesting that the parties “address the constructions of the mean-plus- function limitations in claim 16” and “the impact that a de- termination that the specification of the ’949 patent does not provide adequate corresponding structure for the re- cited functions should have on this proceeding and any fi- nal written decision.” J.A. 5161. After institution, Qualcomm argued in its patent owner response that the means-plus-function terms in claim 16 “do not need to be construed in order for the Board to reach its Final Written Decision” because “[n]one of the argu- ments Qualcomm makes . . . to distinguish the prior art re- quires construction of these limitations,” J.A. 4226, but also argued that constructions Intel had proposed (the same ones Qualcomm had previously proposed in the ITC Case: 20-1828 Document: 98 Page: 9 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 9 proceeding) were sufficient. In reply, Intel did not defend as correct the structure it had identified in its petition; in- stead, it said, “Upon consideration of the Board’s articu- lated concerns, [Intel] agrees that the ’949 specification fails to disclose sufficient structure to perform the recited functions.” J.A. 4325. Intel also agreed with Qualcomm that the Board could address the patentability of claim 16 without construing the means-plus-function limitations. J.A. 4325–26. In sur-reply, Qualcomm pointed out that In- tel had “change[d] its position” about the sufficiency of the structure in the specification; and Qualcomm argued, “[S]hould the Board maintain its position that the specifi- cation does not disclose corresponding structure for the functions, then this precludes the Board from finding that claim 16 is unpatentable.” Patent Owner Sur-Reply at 14– 15, Intel Corp. v. Qualcomm Inc., IPR2018-01334, Paper No. 25 (P.T.A.B. Nov. 8, 2019). In its final written decision, the Board accepted Intel’s position that “‘the ’949 specification fails to disclose suffi- cient structure to perform the recited functions’ for two of the means-plus-function limitations.” Final Written Deci- sion, at *7–8, *26. While acknowledging that both parties agreed that the Board should assess the patentability of claims 16 and 17 even if it concluded that there was insuf- ficient corresponding structure for the functions, the Board concluded, “Because [Intel] has not met its burden under our Rules to show structure corresponding to the claimed function to which we can compare the prior art’s disclosure, we determine [Intel] has not shown, by a preponderance of the evidence, that [claims 16 and 17] are unpatentable un- der 35 U.S.C. § 103(a) as obvious.” Id. at *26. Intel timely appealed. We have statutory jurisdiction under 28 U.S.C. § 1295(a)(4)(A) and 35 U.S.C. §§ 141(c), 319. Qualcomm moved to dismiss the appeal, arguing that Intel lacks Article III standing because Qualcomm had not sued or threatened to sue Intel for infringing the ’949 pa- tent. Apple then moved to intervene in the appeal. We Case: 20-1828 Document: 98 Page: 10 Filed: 12/28/2021 10 INTEL CORPORATION v. QUALCOMM INCORPORATED denied both motions, directing Qualcomm and Intel to ad- dress standing in their merits briefs and permitting Apple to move to file an amicus brief. Apple did so, but its pro- posed brief only repeated its argument that it should be joined as a party, so we denied the motion. II We begin by addressing the jurisdictional standing question, and we arrive at the same conclusion this panel has reached today in Intel Corp. v. Qualcomm Inc., No. 20- 1664 (Fed. Cir. Dec. 28, 2021). Here, as there, Intel has engaged in activity that has already given rise to an in- fringement suit by Qualcomm. J.A. 6214–19 (ITC); J.A. 6239–43 (district court); see also Grit Energy Sols., LLC v. Oren Techs., LLC, 957 F.3d 1309, 1319 (Fed. Cir. 2020). And Qualcomm has not disputed that, in those proceed- ings, it identified an Intel product as the “secondary pro- cessor” of the ’949 patent. Intel Opening Br. 49–50; see also J.A. 6256–57, 6261–62, 6272–73. Thus, for the same rea- sons as in our companion case, “Intel’s risks transcend mere conjecture or hypothesis.” Intel, No. 20-1664, slip op. at 7. That is so even though Intel has only shown that it manufactures the claimed “secondary processor” of the ’949 patent’s claimed inventions, not all the components re- quired by the claims, given the centrality of that compo- nent to the claims, the possibility of direct infringement suits based on product testing, and the possibility of indi- rect infringement suits based on at least inducement. See JTEKT Corp. v. GKN Auto. Ltd., 898 F.3d 1217, 1221 (Fed. Cir. 2018) (“IPR petitioners need not concede infringement to establish standing to appeal.”). We therefore find that Intel has standing and proceed to consider the merits of its appeal. Case: 20-1828 Document: 98 Page: 11 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 11 III A We begin with the claim-construction issue presented respecting claims 1–9 and 12. Claim construction is ulti- mately a question of law, decided de novo on review, as are the intrinsic-evidence aspects of a claim-construction anal- ysis. See, e.g., Data Engine Techs. LLC v. Google LLC, 10 F.4th 1375, 1380 (Fed. Cir. 2021). But we review any un- derlying fact findings about extrinsic evidence, such as ex- tra-patent usage, for substantial-evidence support when the appeal comes from the Board. Compare Monsanto Tech. LLC v. E.I. DuPont de Nemours & Co., 878 F.3d 1336, 1341 (Fed. Cir. 2018) (substantial-evidence review for Board factual findings) with Teva Pharms. USA, Inc. v. Sandoz, Inc., 574 U.S. 318, 331–32 (2015) (clear-error re- view for district court factual findings related to claim con- struction). Here, though there was some expert testimony on the meaning of “hardware buffer,” the Board’s claim- construction reasoning involved only intrinsic evidence from the specification. See Final Written Decision, at *5– 7. We therefore review it de novo. In the inter partes review proceedings before us, brought before November 13, 2018, the Board’s claim-con- struction task was to determine the “broadest reasonable interpretation consistent with the specification.” PPC Broadband, Inc. v. Corning Optical Commc’ns RF, LLC, 815 F.3d 747, 751 (Fed. Cir. 2016); see also Personalized Media Commc’ns, LLC v. Apple Inc., 952 F.3d 1336, 1340 n.2 (Fed. Cir. 2020) (noting the change in Board regula- tions beginning November 13, 2018). When applying that standard, the Board is not limited to choosing the single best interpretation when more than one is reasonable. But it requires that any adopted interpretation be reasonable in light of “general claim construction principles,” which govern in district court. Microsoft Corp. v. Proxyconn, Inc., 789 F.3d 1292, 1298 (Fed. Cir. 2015) (noting that the Board Case: 20-1828 Document: 98 Page: 12 Filed: 12/28/2021 12 INTEL CORPORATION v. QUALCOMM INCORPORATED may not “construe claims during IPR so broadly that its constructions are unreasonable under general claim con- struction principles”), overruled in another respect, Aqua Prods., Inc. v. Matal, 872 F.3d 1290 (Fed. Cir. 2017) (en banc); see also In re Smith Int’l, Inc., 871 F.3d 1375, 1382 (Fed. Cir. 2017). “[T]here is no magic formula or catechism for conduct- ing claim construction.” Phillips v. AWH Corp., 415 F.3d 1303, 1324 (Fed. Cir. 2005) (en banc). Claim language and the specification (written description) are the dominant sources of interpretation, and prosecution history can mat- ter to a lesser degree (though arguments based on prosecu- tion history have not been made here). Id. at 1312–17; see also World Class Tech. Corp. v. Ormco Corp., 769 F.3d 1120, 1123 (Fed. Cir. 2014); Proxyconn, 789 F.3d at 1298 (noting that even under the broadest-reasonable-interpre- tation standard, prosecution history can matter). Of cen- tral importance here, we have explained that “‘[u]ltimately, the interpretation to be given a term can only be deter- mined and confirmed with a full understanding of what the inventors actually invented and intended to envelop with the claim.’” Phillips, 415 F.3d at 1316 (quoting Renishaw plc v. Marposs Societa’ per Azioni, 158 F.3d 1243, 1250 (Fed. Cir. 1998)). Understanding the concrete substance to which the words used in the claim and the intrinsic evi- dence refer is necessary because, “[i]n reviewing the intrin- sic record to construe the claims, we strive to capture the scope of the actual invention, rather than strictly limit the scope of claims to disclosed embodiments or allow the claim language to become divorced from what the specification conveys is the invention.” Retractable Techs., Inc. v. Bec- ton, Dickinson & Co., 653 F.3d 1296, 1305 (Fed. Cir. 2011). Here, it is clear from the claim language that the claim term at issue has meaning, but it is unclear what that meaning is. There is no definition to be found in the intrin- sic evidence. And the determination of that meaning (or range of reasonable meanings) depends on understanding Case: 20-1828 Document: 98 Page: 13 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 13 what the intrinsic evidence makes clear is the substance of the invention—what the inventor “intended to envelop,” Phillips, 415 F.3d at 1316—an understanding that in some cases is usefully clarified by expert testimony (as long as that testimony is consistent with the intrinsic evidence), see id. at 1318. But, we conclude, the Board did not do enough to reach and articulate that understanding, and its claim construction is therefore wanting. The phrase “hardware buffer” appears in claims 1, 2, 8, and 12 of the ’949 patent. We do not discern, and no party has suggested, that “hardware buffer” has a clear, undis- puted meaning in either ordinary English or in relevant technical parlance. Nevertheless, we reach three conclu- sions from the claim language. First, because every buffer in our (physical) world is ul- timately implemented on a physical device (i.e., hardware), a “hardware buffer” must mean something more than just a “buffer implemented in hardware,” as Intel urges, or else the word “hardware” would be erased from the claims. That consequence, while not inevitably disqualifying a con- struction in every patent, is counter to an important prin- ciple of interpretation, for patent claims as for statutes: “It is highly disfavored to construe terms in a way that renders them void, meaningless, or superfluous.” Wasica Finance GmbH v. Continental Automotive Systems, Inc., 853 F.3d 1272, 1288 n.10 (Fed. Cir. 2017); see SimpleAir, Inc. v. Sony Ericsson Mobile Commc’ns AB, 820 F.3d 419, 429 (Fed. Cir. 2016); Stumbo v. Eastman Outdoors, Inc., 508 F.3d 1358, 1362 (Fed. Cir. 2007); Phillips, 415 F.3d at 1314; Duncan v. Walker, 533 U.S. 167, 174 (2001) (refusing to adopt a statutory construction that “would render the word ‘State’ insignificant, if not wholly superfluous”); Salman Ranch Ltd. v. United States, 573 F.3d 1362, 1374 (Fed. Cir. 2009) (“A cardinal rule of statutory construction is that courts should construe statutes so as to avoid rendering superflu- ous any statutory language.” (cleaned up)). Here, the strik- ing fact that Qualcomm, in its claim language, did not just Case: 20-1828 Document: 98 Page: 14 Filed: 12/28/2021 14 INTEL CORPORATION v. QUALCOMM INCORPORATED say “buffer,” but instead said “hardware buffer,” provides a strong reason to avoid the disfavored result of rendering the word “hardware” superfluous. 2 Second, because claim 1 requires both a “system memory” and a “hardware buffer,” there must be some distinction between those two concepts. Third, because claim 2 requires loading the exe- cutable software image “directly from the hardware buffer to the system memory of the secondary processor without copying data between system memory locations on the sec- ondary processor,” the meaning of “hardware buffer” re- lates to the ability to move the software image “directly” to the second processor’s system memory and to avoid “copy- ing data between system memory locations.” Those conclusions from the claim language advance the claim-construction inquiry only so far. And they do not, on their own, provide a concrete basis for a clarifying defini- tion of “hardware buffer.” What is needed, then, is an anal- ysis of the specification to arrive at an understanding of what it teaches about what a “hardware buffer” is, based on both how it uses relevant words and its substantive ex- planations. 3 In this crucial respect, the Board fell short in its analysis here, and we think the Board is better posi- tioned than we are to correct the deficiencies so as to arrive 2 Though Intel asserts that the specification some- times uses the word “hardware” in a redundant manner, Intel Opening Br. at 31, the uses Intel identifies are not instances of modifying the word “buffer,” and Intel does not persuasively show why such uses are actually redundant or, in any event, overcome the fact that the claims (in which relevant readers expect more precise, less discursive draft- ing than in the written description) recite “hardware buffer,” not just “buffer.” 3 To date, the parties have not made substantial claim-construction arguments about the prosecution his- tory. Case: 20-1828 Document: 98 Page: 15 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 15 at the understanding needed to draw a confident conclu- sion about the proper claim construction. We identify some of the deficiencies without suggesting how, if at all, a proper construction will be substantively different from the limited, negative one (excluding “temporary” buffers) that the Board adopted. Although the Board correctly noted that the specifica- tion describes prior art teaching of use of “temporary” buff- ers, it did not explain precisely what “temporary” means or how the patent-described prior-art use relates to the al- leged invention. Final Written Decision, at *7. The Board cited three passages from the ’949 specification (column 2, lines 23–34; column 4, lines 43–47; and column 5, lines 31– 35) that mention the use of a temporary buffer in prior art assertedly being improved by the invention here. But the Board did not analyze exactly how the use of a hardware buffer, as claimed by Qualcomm, would address the con- cerns about the prior-art temporary buffers raised in those passages. Other seemingly significant characteristics of the prior-art buffers (as described in the ’949 patent’s specifi- cation) appear unrelated to their temporal character. Dis- cussing prior art, the specification says that the prior-art buffer “would be some place in system memory.” ’949 pa- tent, col. 2, lines 31–34. That statement aligns with what the claim language already makes clear—Qualcomm’s “hardware buffer” is somehow different from “system memory”—but does not clarify what exactly the difference is. The specification also says that use of a referred-to prior-art buffer would require “copying the data into a tem- porary buffer in the modem processor RAM.” Id., col. 5, lines 31–35. That statement advances the inquiry into the proper claim construction, because it aligns with other specification passages that support an understanding that use of a “hardware buffer” relates to one of the key claimed advances of the invention—the elimination of “extra memory copy operations.” See id., col. 7, line 16 (“Zero Copy Case: 20-1828 Document: 98 Page: 16 Filed: 12/28/2021 16 INTEL CORPORATION v. QUALCOMM INCORPORATED Transport flow” (emphasis added)); id., col. 7, lines 27–30 (“Thus, aspects of the present disclosure avoid extra memory copy operations, thereby improving performance (e.g., reducing the time required to boot secondary proces- sors in a multi-processor system).”); id., col. 9, lines 42–46 (“Accordingly, no extra memory copy operations occur in the secondary processor in the above aspect. Thus, conven- tional techniques employing a temporary buffer for the en- tire image, and the packet header handling, etc., are bypassed in favor of a more efficient direct loading pro- cess.”). As to what the seemingly important relationship be- tween a “hardware buffer” and those characteristics is, we have before us no adequate explanation. The Board’s ex- planation does not clarify the contemplated concrete oper- ations of a “hardware buffer” in this patent. It therefore fails to clarify how a “hardware buffer,” as contemplated in the specification, produces improved efficiency through “di- rect loading” and avoiding “extra memory copy operations.” Id., col. 9, lines 37–46. Nor does the Board’s analysis men- tion the distinctions between “system memory” and “hard- ware buffer” that are drawn both in the claim language and in the specification. What is needed in this case is a more substance-focused analysis than is yet present, in the Board’s opinion or in the present record (at least in the excerpts drawn to our attention), of what the intrinsic evidence shows the as- serted advance to be and how, concretely, the “hardware buffer” relates to that advance. We do not exclude the pos- sibility that the record should be expanded in order to ar- rive at an adequate understanding at the substantive level. The Board’s construction was based on what Qualcomm proposed only in its sur-reply, so the Board did not benefit from expert explanation of technical operations that might bear on the merits of that construction and produce the needed understanding. Case: 20-1828 Document: 98 Page: 17 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 17 The Board’s construction was entirely a negative one—excluding “temporary” buffers. “Although there is no per se rule against negative constructions,” Medicines Co. v. Mylan, Inc., 853 F.3d 1296, 1308 (Fed. Cir. 2017) (cita- tion omitted), which in some cases can be enough to resolve the relevant dispute, the Board’s construction in this case is inadequate. It is not clear what precisely constitutes a “temporary buffer” as recited in the Board’s construction. Compare Intel Opening Br. at 40 (positing that a tempo- rary buffer must be both “allocated or reserved at runtime” and “deallocated to be used for another purpose”), with Qualcomm Resp. Br. at 48 (arguing that a buffer that is “newly allocated each time the system is booted” is tempo- rary). To resolve even that uncertainty requires the kind of additional, substantive understanding discussed above, which seems likely to support an affirmative construction in place of the Board’s purely negative one. Finally, we note two matters that would benefit from attention on remand. In a trial transcript from the Qual- comm-Apple litigation (a transcript that was before the Board here), a named inventor of the ’949 patent testified in some detail about the difficulty of “design[ing] [the] sys- tem so the hardware could place the data right where it needed to be.” Transcript of Jury Trial Day 2 (Vol. 2B) at 222, Intel Corp. v. Qualcomm Inc., IPR2018-01334, Exhibit No. 2004 (P.T.A.B. Dec. 12, 2019). That testimony may bear on the remand inquiry. Additionally, our cases some- times speak of hardware and software implementations of computer functions, reflecting a distinction that appears in usage in the field. See Tomita Techs. USA, LLC v. Nin- tendo Co., 681 F. App’x 967, 972 (Fed. Cir. 2017) (finding that a “software implementation” of an image processing function was not equivalent to a patent’s “hardware imple- mentation” of that function); Texas Digital Systems, Inc. v. Telegenix, Inc., 308 F.3d 1193, 1214–16 (Fed. Cir. 2002) (recognizing the broad range encompassed by a construc- tion as “any firmware, software and/or hardware” that Case: 20-1828 Document: 98 Page: 18 Filed: 12/28/2021 18 INTEL CORPORATION v. QUALCOMM INCORPORATED performs a certain function); Nazomi Commc’ns, Inc. v. ARM Holdings, PLC, 266 F. App’x 935, 936 (Fed. Cir. 2008) (considering a patent on a “hardware accelerator” that “sig- nificantly speed[s] up the processing of Java bytecodes over prior art systems that used software to perform the conver- sion from stack-based instructions to register-based in- structions”). Neither the Board nor the parties have explained what if any bearing that usage may have on a proper understanding of “hardware buffer” in claims 1–9 and 12 here. The Board’s rejection of Intel’s challenge to those claims rested on the determination that the Svensson ref- erence did not disclose the claimed “hardware buffer,” which in turn depended on the Board’s claim construction of that phrase. Because that construction is inadequate and must be reconsidered, we vacate the rejection of Intel’s challenges to claims 1–9 and 12, and we remand for recon- sideration of claim construction as discussed in this opin- ion. B With respect to claims 16 and 17, there is no dispute that claim 16 (and hence dependent claim 17) contains terms that are in means-plus-function format governed by 35 U.S.C. § 112(f). In the IPR2018-01335 proceedings here, after Intel agreed with the Board’s suggestion in the insti- tution decision that two of the means-plus-function terms in claim 16 were indefinite for lack of supporting structure, the Board concluded that Intel’s statement necessarily meant that Intel, as the petitioner, had not met its burden to demonstrate the unpatentability of those claims. See Fi- nal Written Decision, at *26. Intel challenges that merits conclusion. We hold that the Board’s conclusion was error and that a remand is required, because the Board did not decide for itself whether required structure is present in the specification or whether, even if it is not, the absence of Case: 20-1828 Document: 98 Page: 19 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 19 such structure precludes resolution of Intel’s prior-art chal- lenges. Under 35 U.S.C. § 318, as construed in SAS Institute, Inc. v. Iancu, 138 S. Ct. 1348 (2018), the Board is obligated to “issue a final written decision with respect to the patent- ability of” every claim challenged by the petitioner. Id. at 1353. But that obligation does not mean that the Board must reach a determination of the patentability of a claim on the presented prior-art grounds if such a determination is rendered impossible because of the indefiniteness of an essential claim limitation. See Samsung Electronics Amer- ica, Inc. v. Prisua Eng’g Corp., 948 F.3d 1342, 1353 (Fed. Cir. 2020); cf. Anniston Mfg. Co. v. Davis, 301 U.S. 337, 355 (1937) (“We do not think that Congress was attempting to require the impossible.”). In such a case, the statutory es- toppel provision of 35 U.S.C. § 315(e) does not apply, be- cause the problem of indefiniteness is one of the patentee’s own making, not attributable to the challenger. Cochlear Bone Anchored Solutions AB v. Oticon Med. AB, 958 F.3d 1348, 1359 (Fed. Cir. 2020) (citing Samsung, 948 F.3d at 1353 n.3). To avoid confusion going forward, the Board should, in IPRs where that principle applies, clearly state that the final written decision does not include a determi- nation of patentability of any claim that falls within the impossibility category. Importantly, it is not always impossible to adjudicate a prior-art challenge, one way or the other, just because some aspect of a claim renders the claim indefinite. See Sam- sung, 948 F.3d at 1355 (noting that indefiniteness “does not necessarily preclude the Board from addressing the patent- ability of the claims on section 102 and 103 grounds”). For example, the indefiniteness of one limitation may not pre- clude the Board from rejecting a challenge by finding that another limitation is missing from the argued prior art and its argued combinations and modifications. In the other direction, if a claim limitation requires alternative limita- tions A or B, and A is indefinite, but B is not, the Board Case: 20-1828 Document: 98 Page: 20 Filed: 12/28/2021 20 INTEL CORPORATION v. QUALCOMM INCORPORATED may well be able to determine that the argued prior art and its argued combinations or modifications cover the B op- tion, thus satisfying the A or B limitation. See Cochlear, 958 F.3d at 1359–60. The indefiniteness of a limitation (here, a means-plus-function limitation) precludes a pa- tentability determination only when the indefiniteness renders it logically impossible for the Board to reach such a decision. The inter partes review scheme as a whole confirms the limited character of the impossibility qualifier to the SAS obligation. That scheme protects the interests not only of the petitioner in securing a determination on the patenta- bility of a claim, but other interests as well. It protects in- terests of the patentee, as well as of the judicial system and the agency, in diminishing duplication of adjudication bur- dens and risks in specified respects. See, e.g., 35 U.S.C. § 315(b) (timing limit on seeking IPR once district court ac- tion filed), § 315(e) (estoppel provision against other pro- ceedings). And it reflects a public interest in the Board’s answering the patentability questions, reflected in the statutory authorization for the Board to decide an inter partes review even after petitioners have withdrawn. 35 U.S.C. § 317. The Board’s treatment of claims 16 and 17 here was contrary to those principles. The Board simply accepted Intel’s concession that the required structure for the means-plus-function claim was missing. The Board did not itself conclude that the prior-art analysis task was impos- sible, and it could not so conclude here without determining for itself that such structure was missing, a legal question that is part of claim construction. See Noah Systems, Inc. v. Intuit Inc., 675 F.3d 1302, 1311 (Fed. Cir. 2012) (“Whether a claim complies with the definiteness require- ment of 35 U.S.C. § 112 is a matter of claim construction . . . .”). Nor did the Board decide whether it could resolve the patentability dispute even if claim 16 is indefinite. Contrary to Intel’s argument in this court, Intel Opening Case: 20-1828 Document: 98 Page: 21 Filed: 12/28/2021 INTEL CORPORATION v. QUALCOMM INCORPORATED 21 Br. at 45, there is no proper basis at present for not reach- ing the merits of the prior-art challenge to claims 16 and 17 or for declaring inapplicability of estoppel. At the same time, contrary to Qualcomm’s argument in this court, Qual- comm Resp. Br. at 50–51, the Board could not properly en- ter a final written decision on the merits of the prior-art challenges with estoppel effect, as the Board did not assess those merits. On remand, the Board must decide one or both of two issues. One is whether it can resolve the prior-art chal- lenge to the patentability of claims 16 and 17 despite the potential indefiniteness of the means-plus-function terms. The other is whether those terms are actually indefinite. We express no view on which issue the Board should con- sider first. If the Board determines both that there is in- definiteness and that such indefiniteness renders it impossible to adjudicate the prior-art challenge on its mer- its, then the Board should conclude that it is impossible to reach a decision on the merits of the challenge and so state in its decision. See Samsung, 948 F.3d at 1358. IV For the foregoing reasons, we vacate the Board’s con- struction of the term “hardware buffer,” its determination that claims 1–9 and 12 were non-obvious over the prior art, and its conclusion that claims 16–17 lacked sufficient cor- responding structure in the specification, and we remand for further proceedings consistent with this opinion. The parties shall bear their own costs. VACATED AND REMANDED