Samsung Electronics Co., Ltd. v. Elm 3ds Innovations, LLC

United States Court of Appeals for the Federal Circuit ______________________ SAMSUNG ELECTRONICS CO., LTD., MICRON TECHNOLOGY, INC., SK HYNIX INC., Appellants v. ELM 3DS INNOVATIONS, LLC, Appellee ______________________ 2017-2474, 2017-2475, 2017-2476, 2017-2478, 2017-2479, 2017-2480, 2017-2482, 2017-2483, 2018-1050, 2018-1079, 2018-1080, 2018-1081, 2018-1082 ______________________ Appeals from the United States Patent and Trademark Office, Patent Trial and Appeal Board in Nos. IPR2016- 00386, IPR2016-00387, IPR2016-00388, IPR2016-00390, IPR2016-00391, IPR2016-00393, IPR2016-00394, IPR2016-00395, IPR2016-00687, IPR2016-00691, IPR2016-00708, IPR2016-00770, IPR2016-00786. ______________________ Decided: June 12, 2019 ______________________ RUFFIN B. CORDELL, Fish & Richardson PC, Washing- ton, DC, argued for all appellants. Appellants Micron Technology, Inc., SK Hynix Inc. also represented by CHRISTOPHER DRYER, TIMOTHY W. RIFFE, ROBERT ANDREW SCHWENTKER, ADAM SHARTZER; CRAIG E. COUNTRYMAN, RYAN LYNN FREI, OLIVER RICHARDS, San Diego, CA. 2 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC NAVEEN MODI, Paul Hastings LLP, Washington, DC, for appellant Samsung Electronics Co., Ltd. Also repre- sented by PHILLIP W. CITROEN, ALLAN SOOBERT. WILLIAM MEUNIER, Mintz, Levin, Cohn, Ferris, Glovsky and Popeo, P.C., Boston, MA, argued for appellee. Also rep- resented by KEVIN AMENDT, SANDRA BADIN, MATTHEW STEPHEN GALICA, MICHAEL NEWMAN, MICHAEL TIMOTHY RENAUD, JAMES M. WODARSKI. ______________________ Before MOORE, REYNA, and CHEN, Circuit Judges. MOORE, Circuit Judge. Samsung Electronics Co., Ltd., Micron Technology, Inc., and SK Hynix Inc. (collectively, “Petitioners”) appeal from the final written decisions of the Patent Trial and Ap- peal Board in thirteen inter partes reviews holding that they did not establish the unpatentability of 105 claims across eleven patents (“Challenged Patents”). Given that each challenged claim requires a low-tensile-stress dielec- tric, and substantial evidence supports the Board’s finding that a person of ordinary skill in the art would not have reasonably expected success in combining the prior art to meet this limitation, we affirm. SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 3 LLC BACKGROUND Appellee Elm 3DS Innovations LLC (“Elm”) is the owner of the Challenged Patents, 1 which share a specifica- tion and all relate to “stacked integrated circuit memory.” 2 ’672 patent at 1:7–8. The Challenged Patents are the sub- ject of co-pending litigation between Elm and Petitioners. The Board instituted inter partes review based on thir- teen petitions filed by Petitioners. Among others not at is- sue on appeal, the petitions challenged the following claims: claims 17–18, 22, 84, 95, 129–32, 145–46, and 152 of the ’672 patent (IPR2016-00386); claims 1–2, 8, 14, 31– 32, 44, 46, and 52–54 of the ’778 patent (IPR2016-00387); claims 10–12, 18–20, 60–63, 67, 70–73, and 77 of the ’239 patent (IPR2016-00388 and IPR2016-00393); claims 1–3, 30–31, 33, 40–41, and 44 of the ’542 patent (IPR2016- 00390); claims 30, 34, 36, 135–138, and 147 of the ’862 pa- tent (IPR2016-00391); claims 36 and 51 of the ’617 patent (IPR2016-00394); claims 1, 10–11, and 13–14 of the ’732 patent (IPR2016-00395); claims 1, 7, 17–18, and 33 of the ’119 patent (IPR2016-00687); claims 1 and 20–23 of the ’004 patent (IPR2016-00691); claims 1, 12–13, 24, 36–38, 53, 83, 86–87, and 132 of the ’499 patent (IPR2016-00708 and IPR2016-00770); and claims 58, 60–61, and 67 of the ’570 patent (IPR2016-00786). Each ground challenging the claims was based on obviousness and asserted either U.S. Patent No. 5,202,754 (“Bertin”) or a 1996 article by Kee-Ho Yu, et. al., titled “Real-Time Microvision System with Three-Dimensional Integration Structure” (“Yu”) as the 1 The patents at issue are U.S. Patent Nos. 8,653,672; 8,841,778; 7,193,239; 8,629,542; 8,796,862; 8,410,617; 7,504,732; 8,928,119; 7,474,004; 8,907,499; and 8,933,570. 2 For simplicity, this opinion cites only to the specifi- cation of the ’672 patent. 4 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC primary reference in combination with, relevant here, U.S. Patent No. 5,354,695 (“Leedy”). 3 The Board held that Petitioners had not met their bur- den of demonstrating that the claims were unpatentable. Specifically, it found that the prior art did not disclose the “substantially flexible” limitation. It also found that Peti- tioners did not demonstrate a motivation to combine Bertin or Yu with Leedy or a reasonable expectation of success in doing so. Petitioners timely filed notices of appeal, and the appeals were consolidated. We have jurisdiction pursuant to 28 U.S.C. § 1295(a)(4)(A). DISCUSSION I. Claim Construction “We review the Board’s constructions based on intrin- sic evidence de novo and its factual findings based on ex- trinsic evidence for substantial evidence.” HTC Corp. v. Cellular Commc’ns Equip., LLC, 877 F.3d 1361, 1367 (Fed. Cir. 2017). The Board construes claims in an unexpired patent according to their broadest reasonable interpreta- tion in light of the specification. 37 C.F.R. § 42.100(b) (2017). 4 Claims of an expired patent are construed accord- ing to the standard applied by district courts. See In re CSB-Sys. Int’l, Inc., 832 F.3d 1335, 1341 (Fed. Cir. 2016) (referencing Phillips v. AWH Corp., 415 F.3d 1303 (Fed. 3 Claim 1 of the ’499 patent was challenged based on U.S. Patent No. 5,731,945, which contains the same disclo- sure as Bertin and adds details not relevant to this appeal. 4 The Board’s decisions issued prior to the effective date of the U.S. Patent and Trademark Office’s change to the claim construction standard applied in inter partes re- view. See Changes to the Claim Construction Standard for Interpreting Claims in Trial Proceedings Before the Patent Trial and Appeal Board, 83 Fed. Reg. 51,340 (Oct. 11, 2018). SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 5 LLC Cir. 2005) (en banc)). While some patents were expired at the time of the Board’s final written decision and others were not, the parties agree that the different claim con- struction standards do not impact the outcome. Appel- lants’ Br. 44; Appellee’s Br. 41. The parties have not contested the Board’s application of the Phillips claim con- struction standard. All challenged claims except for claims 1 and 14 of the ’778 patent use “substantially flexible” in at least one of two ways. The first is to modify the term “semiconductor sub- strate.” Claim 129 of the ’672 patent illustrates the use in this context (emphasis added): An integrated circuit structure comprising: a first substrate comprising a first surface supporting interconnect contacts; a substantially flexible semiconductor sec- ond substrate comprising a first surface and a second surface at least one of which supports interconnect contacts, wherein the second surface is opposite the first sur- face and wherein the second surface of the second substrate is formed by removal of semiconductor material from the second substrate and is smoothed or polished after removal of the semiconductor material; and conductive paths between the interconnect contacts supported by the first surface of the first substrate and of the interconnect contacts supported by the second sub- strate; wherein the first substrate and the second substrate overlap fully or partially in a stacked relationship; and 6 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC wherein the integrated circuit structure further comprises a low-stress silicon- based dielectric material having a stress of 5×108 dynes/cm2 tensile or less. “Substantially flexible” is also used to modify “circuit lay- ers,” and other similar terms. 5 Claim 30 of the ’862 patent illustrates how “substantially flexible” is used in this con- text (emphasis added): A stacked circuit structure comprising: a plurality of stacked, thin, substantially flexible circuit layers at least one of which comprises a thinned, substantially flexible monocrystalline semiconductor substrate of one piece; wherein at least one of the substantially flexible circuit layers comprises at least one memory array comprising memory cells and a low stress silicon-based dielectric material; and at least one vertical interconnection that passes through at least one of the plurality of stacked, thin, substantially flexible cir- cuit layers. 5 See, e.g., ’239 patent at Claim 60 (“substantially flexible” die); ’004 patent at Claim 1 (“substantially flexible integrated circuits”); ’732 patent at Claim 1 (“substantially flexible integrated circuit layer”). The parties do not treat this difference in terminology as affecting the construction of “substantially flexible.” Accordingly, our construction of “substantially flexible” applies across all its uses. SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 7 LLC In each context, the Board relied on a general-purpose dic- tionary to construe “substantially flexible” to mean “largely able to bend without breaking.” E.g., J.A. 31. “Claim terms generally are construed in accordance with the ordinary and customary meaning they would have to one of ordinary skill in the art in light of the specification and the prosecution history.” Aventis Pharma S.A. v. Hos- pira, Inc., 675 F.3d 1324, 1329 (Fed. Cir. 2012) (citing Phil- lips, 415 F.3d at 1312). Extrinsic evidence may also be considered in construing a claim, though “it is less signifi- cant than the intrinsic record in determining the legally operative meaning of claim language.” Phillips, 415 F.3d at 1317 (internal quotation marks omitted). We will devi- ate from a claim term’s ordinary meaning “when a patentee sets out a definition and acts as its own lexicographer” or “when the patentee disavows the full scope of a claim term either in the specification or during prosecution.” Aventis, 675 F.3d at 1330 (quoting Thorner v. Sony Computer Entm’t Am. L.L.C., 669 F.3d 1362, 1365 (Fed. Cir. 2012)). The parties dispute the meaning of “substantially flex- ible.” “Where multiple patents derive from the same par- ent application and share many common terms, we must interpret the claims consistently across all asserted pa- tents.” SightSound Techs., LLC v. Apple Inc., 809 F.3d 1307, 1316 (Fed. Cir. 2015) (internal quotation marks omit- ted). The parties do not argue that the definition of “sub- stantially flexible” depends on the patent or claim in which it is used. Because the Challenged Patents derive from the same parent application and use “substantially flexible” throughout, we construe that term the same way for each Challenged Patent. Petitioners argue the intrinsic record supports a con- struction of “substantially flexible” substrate as a “sub- strate that has been thinned to a thickness of less than 50 µm and subsequently polished or smoothed.” Appellants’ 8 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC Br. 36. Specifically, they rely on the specification’s disclo- sure of step “2A” in a fabrication sequence for a “3DS memory circuit,” which states: “Grind the backside or ex- posed surface of the second circuit substrate to a thickness of less than 50 µm and then polish or smooth the surface. The thinned substrate is now a substantially flexible sub- strate.” ’672 patent at 9:3–6; see also id. at 2:66–67, 3:5–8 (stating that a feature of the stacked circuit assembly tech- nology includes “[t]hinning of the memory circuit to less than about 50 µm in thickness forming a substantially flex- ible substrate”). Though these disclosures refer to the sub- strate being substantially flexible, Petitioners argue they apply with equal force to the claims reciting “substantially flexible” circuit layers, and similar limitations, because the prosecution history requires that a substantially flexible circuit layer includes a substantially flexible substrate. Elm responds that the Board’s construction is con- sistent with the ordinary meaning of “substantially flexi- ble” and the specification’s distinction between flexible and rigid substrates. It criticizes Petitioners’ proposed con- struction as departing from the ordinary meaning, since the flexibility of a material depends on more than how thin and polished it is. Citing the declaration of Petitioners’ ex- pert Dr. Paul Franzon, Elm argues the flexibility of a sem- iconductor substrate depends on the substrate’s elastic modulus, crystal orientation, and dimensions. Appellee’s Br. 48–49 (citing J.A. 2191–92 ¶ 71). Neither party’s construction is quite right. We begin our analysis with the claim language. The claims indicate that, at least in some situations, thinning and polishing a substrate is one way of forming a substantially flexible sub- strate. For example, claim 31 of the ’778 patent recites “the semiconductor substrate is thinned and polished or smoothed such that the semiconductor substrate is sub- stantially flexible.” See also ’862 patent at Claim 147 (re- citing “the polished or smoothed backside [of a thinned, monocrystalline semiconductor substrate] enables the . . . SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 9 LLC substrate to be substantially flexible, and the polished or smoothed backside reduces the vulnerability of the . . . sub- strate to fracture as a result of flexing”). But that does not mean this is the only way to achieve substantial flexibility. The claim on which claim 31 depends recites “the semicon- ductor substrate is substantially flexible,” ’778 patent at Claim 2, implying that it covers substantially flexible sub- strates formed in ways other than the one recited in claim 31, Clearstream Wastewater Sys., Inc. v. Hydro-Action, Inc., 206 F.3d 1440, 1446 (Fed. Cir. 2000) (“Under the doc- trine of claim differentiation, it is presumed that different words used in different claims result in a difference in meaning and scope for each of the claims.”). Claim 51 of the ’617 patent recites “the bottomside of the first substrate is polished to make the substrate substantially flexible,” with no specific “thinned” limitation. Conversely, claim 8 of the ’778 patent lacks a polishing limitation, reciting a substrate that “is formed from a semiconductor wafer and is thinned and substantially flexible.” The claims alone do not support limiting “substantially flexible” to Petitioners’ proposed construction. The prosecution history, on the other hand, shows that “substantially flexible” is narrower than the Board’s con- struction of “largely able to bend without breaking.” E.g., J.A. 31. During prosecution of the application that led to the ’499 patent, the examiner objected to the use of the term “substantially flexible” because it rendered the claim’s scope unclear. J.A. 10260. Elm responded that “the meaning of [substantially flexible] as used in the claims is clearly explained in the specification,” citing to step 2A in the fabrication sequence. J.A. 10275. “As described in this passage,” Elm continued, “a semiconductor substrate is caused to be substantially flexible by thinning it to 50 mi- crons or less and polishing or smoothing the thinned semi- conductor substrate to relieve stress. The phrase ‘substantially flexible’ is used in the claims consistent with this description, which is unambiguous.” Id. To overcome 10 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC the examiner’s objection, Elm clearly and unambiguously disclaimed claim scope. For a semiconductor substrate to be “substantially flexible” according to the claims, it must be thinned to 50 microns or less and polished or smoothed. This definition of “substantially flexible” applies to all its uses. In response to a rejection of claims reciting a sub- stantially flexible circuit layer in an application related to the Challenged Patents, Elm stated that “a substantially flexible semiconductor substrate is a necessary but not a sufficient condition for a substantially flexible circuit layer.” J.A. 10316 (emphasis in original). Reinforcing this point, Elm in a response involving another related applica- tion explained: Two features are required to achieve substantial flexibility. One is that the semiconductor material must be sufficiently thin, e.g., 50 microns or less. . . . The other is that the dielectric material used in processing the semiconductor material must be sufficiently low stress. Otherwise, sub- stantial flexibility is defeated. As set forth in the present specification, stress of 5 x 108 dynes/cm2 or less has been demonstrated to satisfy this require- ment. J.A. 16038 (emphasis added). See also J.A. 10314 (“[A] cir- cuit layer requires one or more dielectric layers. . . . For a circuit layer to be substantially flexible, Applicant has found that the dielectric material must have low tensile stress, for example, 5 x 108 dynes/cm2 tensile.”). Consid- ered in its entirety, the prosecution history clearly and un- ambiguously demonstrates that a substantially flexible circuit layer, and similar terms, must contain a substan- tially flexible semiconductor substrate and a sufficiently low tensile stress dielectric material. We see nothing in the specification or prosecution history that limits the dielec- tric to a particular stress value. Both merely provide as an SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 11 LLC example that a tensile stress of 5 x 108 dynes/cm2 is suffi- cient. This is not, however, the end of the construction. The prosecution history makes clear that “substantially flexi- ble” cannot be read to cover rigid substrates and circuit lay- ers. See J.A. 15397 (criticizing the prior art substrate because it is “rigid”); J.A. 16039 (stating the prior art “de- scribe[s] a stacked integrated circuit formed on a rigid car- rier . . . , suggesting that the stacked integrated circuit is in fact inflexible” (emphasis in original)). Based on expert testimony from Dr. Franzon, the Board found that “there are a number of factors that, within the context of semicon- ductor processing, determine the flexibility of a semicon- ductor substrate,” including the type of semiconductor substrate, the crystal orientation of the material, and the physical dimensions of the substrate. E.g., J.A. 27 (citing J.A. 2191–92 ¶ 71). This suggests thinning the semicon- ductor substrate to 50 µm and subsequently polishing or smoothing it is necessary but not necessarily sufficient to make the substrate substantially flexible. To ensure that the construction of “substantially flexible” cannot be read to cover a rigid substrate or circuit layer, we interpret a substantially flexible semiconductor substrate as a semi- conductor substrate that is thinned to 50 µm and subse- quently polished or smoothed such that it is largely able to bend without breaking. Likewise, we interpret a substan- tially flexible circuit layer as a circuit layer that is largely able to bend without breaking and contains a substantially flexible semiconductor substrate and a sufficiently low ten- sile stress dielectric material. II. Obviousness We review the Board’s legal determinations de novo and its underlying factual findings for substantial evi- dence. Belden Inc. v. Berk-Tek LLC, 805 F.3d 1064, 1073 (Fed. Cir. 2013). Obviousness is a question of law based on underlying facts. Id. Whether there was a motivation to 12 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC combine references and a reasonable expectation of success in doing so to meet the limitations of the claimed invention are questions of fact. Intelligent Bio-Sys., Inc. v. Illumina Cambridge Ltd., 821 F.3d 1359, 1367–68 (Fed. Cir. 2016). Each ground of unpatentability relied on either Bertin or Yu in combination with Leedy, along with other refer- ences not relevant on appeal. Bertin discloses “[a] fabrica- tion method and resultant three-dimensional multichip package having a densely stacked array of semiconductor chips interconnected at least partially by means of a plu- rality of metallized trenches.” J.A. 1206 at Abstract. “[P]rocessing begins with a semiconductor device 50 (pref- erably comprising a wafer) having a substrate 52 and an active layer 54, which is typically positioned at least par- tially therein.” J.A. 1216 at 3:50–53. A dielectric layer is grown over the active layer. Id. at 3:60–62. 6 Yu discloses a fabrication process for a 3D integration structure in which a silicon wafer is glued to quartz glass, thinned and polished, and bonded to a thick wafer. The structure in- cludes a “field oxide,” depicted in two figures as silicon di- oxide. J.A. 1350. Leedy discloses a method of fabricating “integrated circuits from flexible membranes formed of very thin low stress dielectric materials, such as silicon di- oxide or silicon nitride, and semiconductor layers.” J.A. 1229 at Abstract. Regarding the Bertin-Leedy combinations, Petitioners proposed depositing a low-stress dielectric material using plasma-enhanced chemical vapor deposition (“PECVD”), as disclosed in Leedy, instead of growing the dielectric layer, as disclosed in Bertin. The Board found that a person of ordinary skill in the art would not have been motivated to make such a combination and would not have had a rea- sonable expectation of success in doing so. It credited the 6 A dielectric is an insulator used in electric circuits. J.A. 2375 ¶ 33. SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 13 LLC testimony of Elm’s expert Dr. Alexander Glew that PECVD was incompatible with Bertin’s integrated circuit. Given the complexity involved in integrated circuit fabrication, it found Dr. Franzon’s testimony that PECVD had certain benefits and that Leedy and Bertin are in the same techno- logical field was insufficient to meet Petitioners’ burden. As a result, it found Petitioners failed to adequately ex- plain “how [Bertin’s] fabrication process would be changed to use [Leedy’s] dielectric material, which is formed in a quite different manner than [Bertin’s] dielectric layer.” J.A. 77. The Board’s finding as to a lack of reasonable ex- pectation of success is supported by substantial evidence. Bertin discloses that “[a] dielectric layer 60, for exam- ple, [silicon dioxide], is grown over active layer 54 of device 50.” J.A. 1216 at 3:60–62. Dr. Glew testified that a silicon dioxide dielectric that is grown directly over circuit compo- nents must be high-purity to not damage the circuit com- ponents. J.A. 2415 ¶ 128. As a result, one of ordinary skill in the art would have known from Bertin’s description that the dielectric layer 60 “was grown at high temperatures us- ing thermal oxidation.” J.A. 2415–16 ¶ 128; see also J.A. 1527 (acknowledging in the Petition that Bertin dis- closes “thermally grown oxides”). Thermal oxidation is a process in which silicon at the surface of a wafer is con- verted to high-purity silicon dioxide by exposing it to oxy- gen at high temperatures, typically between 900 °C and 1200 °C. J.A. 2387–88 ¶¶ 66–67. Substantial evidence supports the Board’s finding that Petitioners did not adequately explain how Bertin’s fabri- cation process would be changed to use Leedy’s dielectric material. The Petition asserted that Leedy’s dielectric ma- terial could “easily be used in place of” Bertin’s dielectric using PECVD. J.A. 1527. In support of this argument, Dr. Franzon testified that PECVD “was a commonly avail- able deposition technique that could have been used in place of” Bertin’s technique for growing dielectrics. J.A. 2207 ¶ 101. He also testified that Leedy explains that 14 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC “its dielectric deposition processes are compatible with con- ventional integrated circuit fabrication methods.” J.A. 2206–07 ¶ 101. For example, Leedy states that “[t]he dielectric membrane is compatible with most higher tem- perature [integrated circuit] processing techniques.” J.A. 1296 at 5:32–33. Evidence shows that selecting a dielectric and a method of forming that dielectric is more complicated than Petitioners suggest. A specific dielectric, like silicon diox- ide, can have “vastly different characteristics and behav- iors” depending on how it is made. J.A. 2386 ¶ 63. Dr. Glew identified eighteen factors to be considered when selecting a dielectric and method of formation. Those fac- tors include: (1) dielectric constant, (2) breakdown field strength, (3) leakage, (4) surface conductance, (5) moisture absorption or permeability to mois- ture, (6) stress, (7) adhesion to aluminum, (8) ad- hesion to dielectric layers above or below, (9) stability, (10) etch rate, (11) permeability to hy- drogen, (12) amount of incorporated electrical charge or dipoles, (13) amount of impurities, (14) quality of step coverage, (15) the thickness and uniformity of the film, (16) ability to provide good doped uniformity across a wafer, (17) defect den- sity, [and] (18) amount of residual constituents that outgas during later processing. J.A. 2421 ¶ 139. Dr. Glew stated that most of these factors are unknown here with respect to Leedy’s dielectric, so a person of ordinary skill in the art could not conclude that it would have been obvious to make the proposed substitu- tion. In light of the complexity of semiconductor fabrica- tion, the Board found Petitioners’ explanation lacking. The Board’s finding that PECVD is “quite different” from thermal oxidation is supported by substantial evi- SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 15 LLC dence. J.A. 77. As the name suggests, PECVD is a deposi- tion process, unlike thermal oxidation, which is a growth process. PECVD is performed at 400 °C or less and uses plasma to create a reaction between the surface of a wafer and chemical vapors that include the atoms or molecules to be deposited. In contrast to thermal oxidation, which yields a high-purity dielectric, Dr. Glew testified that die- lectrics deposited using PECVD “include impurities that make them unusable for a variety of applications requiring higher purity.” J.A. 2392 ¶ 77. According to Dr. Glew, this creates a problem when attempting to implement Leedy’s dielectric into Bertin using PECVD because the dielectric layer of Bertin must be highly pure to not damage the cir- cuit components. J.A. 2415–16 ¶ 128. The dielectric pro- duced using PECVD would not be sufficiently pure. J.A. 2416 ¶ 130. He also testified that PECVD “cannot be used because positive ions present in the plasma can strike and damage the wafer and the exposed active components in and on its surface.” J.A. 2423 ¶ 142. Petitioners argue the Board erred when it declined to resolve a dispute about front-end-of-line and back-end-of- line processing steps, especially when it relied on Dr. Glew’s testimony that assumed Bertin’s dielectric was grown during the front-end-of-line phase of the fabrication process. Dr. Glew’s testimony was that if Leedy’s dielectric replaced Bertin’s at the same phase in the fabrication pro- cess, PECVD could not be used “because the resulting die- lectric would not (1) be sufficiently pure; (2) have the ability to adhere sufficiently to the semiconductor wafer; and (3) be able to withstand high temperatures of the re- maining [front-end-of-line] steps,” which generally occur at higher temperatures than the back-end-of-line steps, “without changing its form.” J.A. 2422–23 ¶ 142. We see no legal error in the Board’s decision. First, the Board found that even assuming Petitioners’ contentions were ac- curate, their explanation was lacking. Second, we under- stand the Board’s opinion as finding it unnecessary to 16 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC decide this issue because, at least as to Dr. Glew’s first two points, the timing would not matter. Though Petitioners disputed these facts in their Reply below, they did so based on attorney argument without premising that argument on the timing of applying PECVD. J.A. 1811–12. Moreover, “[t]he possibility of drawing two inconsistent conclusions from the evidence does not prevent an administrative agency’s finding from being supported by substantial evi- dence.” In re Applied Materials, Inc., 692 F.3d 1289, 1294 (Fed. Cir. 2012). Petitioners also argue the Board improperly required proof that unclaimed elements were combinable. “It is well-established that a determination of obviousness based on teachings from multiple references does not require an actual, physical substitution of elements.” In re Mouttet, 686 F.3d 1322, 1332 (Fed. Cir. 2012). “What matters in the § 103 nonobviousness determination is whether a person of ordinary skill in the art, having all the teachings of the ref- erences before him, is able to produce the structure defined by the claim.” Orthopedic Equip. Co., Inc. v. United States, 702 F.2d 1005, 1013 (Fed. Cir. 1983). The Board did not require unclaimed elements be combinable. Rather, it re- peatedly stated that integrated-circuit technology is com- plex and, as such, looked for specific evidence that a person of ordinary skill in the art would have reasonably expected success in combining Bertin’s fabrication process and Leedy’s dielectric material. Petitioners specifically argued in its Petition that “PECVD . . . could have been used in place of the dielectric growing techniques described in Ber- tin to obtain the predictable result of stacked [integrated circuits] having low tensile stress dielectrics.” J.A. 1528. The Board ultimately determined that Petitioners’ evi- dence in support of that combination was insufficient. We will not fault the Board for analyzing Petitioners’ obvious- ness grounds in the way presented in the Petition. SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, 17 LLC Finally, Petitioners argue there was a reasonable ex- pectation of success because the Challenged Patents incor- porate Leedy by reference. The patents state that “dielectrics in low stress . . . such as low stress silicon diox- ide and silicon nitride . . . are discussed at length in [Leedy], incorporated herein by reference.” ’672 patent at 8:46–53. Petitioners argue that the failure to mention any technical problems with using Leedy’s dielectrics indicates that doing so was trivial. The Board considered this argu- ment and rejected it. We find the Petitioners’ argument too speculative to warrant a conclusion that the Board’s fac- tual finding lacked substantial evidence. The arguments related to the Yu-Leedy combinations were substantially similar to the Bertin-Leedy combina- tions. According to the Petition, it would have been obvious to replace Yu’s silicon dioxide and processes for forming it with the dielectric and deposition process taught by Leedy. “Using [Leedy’s] dielectric materials and deposition tech- niques in the manufacture of Yu’s 3D LSI results in” the combination disclosing the low-tensile-stress-dielectric limitation. J.A. 1558. Dr. Franzen’s testimony in support of this combination was identical to the combination in the Bertin-Leedy grounds. See J.A. 2206–08 ¶¶ 99–103. The Board found that Petitioners failed to meet their burden for substantially the same reasons. The evidence discussed as to why a person of ordinary skill in the art would not have reasonably expected success in making the proposed combination applies equally here. Dr. Glew testified that Yu identifies its dielectric as a “field oxide,” which one of ordinary skill in the art would have understood is a highly pure dielectric grown directly on the silicon substrate at high temperatures using thermal oxi- dation. J.A. 2418–19 ¶¶ 134–35 (citing J.A. 1350). His tes- timony about why a person of ordinary skill in the art would not have reasonably expected success using PECVD to deposit Leedy’s dielectric was likewise the same. Peti- tioners raise no argument on appeal that distinguishes the 18 SAMSUNG ELECTRONICS CO., LTD. v. ELM 3DS INNOVATIONS, LLC Bertin-Leedy grounds from the Yu-Leedy grounds. Sub- stantial evidence supports the Board’s finding of a lack of reasonable expectation of success. This issue is dispositive as to all challenged claims. All claims except claims 60, 67, 70, and 77 of the ’239 patent; claims 1 and 44 of the ’542 patent; claim 1 of the ’119 pa- tent; and claim 58 of the ’570 patent explicitly require a low tensile stress dielectric. These claims recite either a sub- stantially flexible die or integrated circuit, meaning they too require a low tensile stress dielectric under the proper claim construction. We thus affirm the Board’s finding as to a lack of reasonable expectation of success and need not reach Petitioners’ remaining arguments. CONCLUSION Because we hold that substantial evidence supports the Board’s finding of a lack of reasonable expectation of success, we need not address the Board’s separate findings that the prior art does not teach the “substantially flexible” limitation or that a person of ordinary skill in the art would have lacked a motivation to combine. For the foregoing reasons, we affirm. AFFIRMED