Vlsi Technology LLC v. Intel Corporation

Case: 22-1906    Document: 72    Page: 1    Filed: 12/04/2023




   United States Court of Appeals
       for the Federal Circuit
                  ______________________

                VLSI TECHNOLOGY LLC,
                     Plaintiff-Appellee

                            v.

                INTEL CORPORATION,
                  Defendant-Appellant
                 ______________________

                        2022-1906
                  ______________________

    Appeal from the United States District Court for the
 Western District of Texas in No. 6:21-cv-00057-ADA, Judge
 Alan D. Albright.
                   ______________________

                Decided: December 4, 2023
                 ______________________

    JEFFREY A. LAMKEN, MoloLamken LLP, Washington,
 DC, argued for plaintiff-appellee. Also represented by
 RAYINER HASHEM, MICHAEL GREGORY PATTILLO, JR.;
 MORGAN CHU, BENJAMIN W. HATTENBACH, ALAN J.
 HEINRICH, AMY E. PROCTOR, DOMINIK SLUSARCZYK,
 CHARLOTTE J. WEN, Irell & Manella LLP, Los Angeles, CA;
 BABAK REDJAIAN, Newport Beach, CA.

     WILLIAM F. LEE, Wilmer Cutler Pickering Hale and
 Dorr LLP, Boston, MA, argued for defendant-appellant.
 Also represented by ALISON BURTON, LAUREN B. FLETCHER,
 JOSEPH J. MUELLER; STEVEN JARED HORN, AMANDA L.
Case: 22-1906    Document: 72      Page: 2    Filed: 12/04/2023




 2                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 MAJOR, Washington, DC; MARY VIRGINIA SOOTER, Denver,
 CO.
               ______________________

     Before LOURIE, DYK, and TARANTO, Circuit Judges.
 TARANTO, Circuit Judge.
      VLSI Technology LLC owns 

U.S. Patent No. 7,523,373,

 titled “Minimum Memory Operating Voltage Technique”
 and 

U.S. Patent No. 7,725,759,

 titled “System and Method
 of Managing Clock Speed in an Electronic Device.” VLSI
 sued Intel Corporation, alleging infringement of both pa-
 tents, and after a trial, the jury found infringement of both
 patents and awarded separate damages for each. The dis-
 trict court then denied Intel’s post-trial motions on various
 issues concerning infringement and damages. It simulta-
 neously denied Intel’s pre-trial motion seeking to add a li-
 cense defense to the case and to sever that defense from the
 rest of the case and stay its adjudication.
      Intel appeals. We affirm the judgment of infringement
 of the ʼ373 patent but reverse the judgment of infringement
 of the ʼ759 patent. We vacate the award of damages for the
 ʼ373 patent and remand for a new trial limited to damages.
 We reverse the denial of the motion for leave to amend to
 add the license defense.
                               I
     On April 11, 2019, VLSI sued Intel for patent infringe-
 ment. VLSI asserted claims 1, 5, 6, 9, and 11 of the ʼ373
 patent and claims 14, 17, 18, and 24 of the ʼ759 patent. Af-
 ter a six-day trial, the jury found that Intel literally in-
 fringed all asserted claims of the ʼ373 patent and that Intel
 infringed all asserted claims of the ʼ759 patent, but only
 under the doctrine of equivalents.
Case: 22-1906    Document: 72     Page: 3   Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                 3



                             A
     The ’373 patent describes, among other things, a fea-
 tured embodiment in which an integrated circuit has a
 memory and a processor; the memory has a minimum op-
 erating voltage; and when the processor is provided power
 at a voltage below the memory-minimum level (e.g., when
 the processor is in a low-power state), the memory is pro-
 vided power at a higher voltage than the processor. ʼ373
 patent, Abstract. Figure 1 illustrates the circuit of that
 embodiment:




     Figure 1 discloses a memory 18, which has a minimum
 operating voltage. 

Id.,

 col. 6, lines 33–36. Figure 1 also
 discloses two voltage regulators: voltage regulator 24,
 which provides a scalable power supply voltage, VDDlogic,
 to both processor 16 and memory 18; and voltage regulator
 26, which provides a substantially fixed power supply volt-
 age, VDDmem, just to memory 18. 

Id.,

 col. 3, lines 21–29.
 The memory 18 includes a power supply selector 21, which
 receives both power supply voltages VDDmem and VDD-
 logic, and provides one of them to memory array 22 as the
 memory operating voltage. 

Id.,

 col. 2, lines 50–57. “In one
 embodiment, while VDDlogic remains above a minimum
 operating voltage required for successful reads of memory
 array 22, power supply selector 21 selects VDDlogic as the
Case: 22-1906     Document: 72     Page: 4    Filed: 12/04/2023




 4                 VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 memory operating voltage provided to memory array 22
 . . . .” 

Id.,

 col. 3, lines 30–35. “When VDDlogic is scaled to
 a voltage that is below the minimum memory operating
 voltage required for reads, power supply selector 21 selects
 the higher voltage, VDDmem . . . .” 

Id.,

 col. 3, lines 35–39.
     The claims are not limited to the featured embodiment
 just described. Independent claim 1 claims a method:
     1. A method, comprising:
         providing an integrated circuit with a
         memory;
         operating the memory with an operating
         voltage;
         determining a value of a minimum operat-
         ing voltage of the memory;
         providing a non-volatile memory (NVM) lo-
         cation;
         storing the value of the minimum operat-
         ing voltage of the memory in the NVM lo-
         cation;
         providing a functional circuit on the inte-
         grated circuit exclusive of the memory;
         providing a first regulated voltage to the
         functional circuit;
         providing a second regulated voltage, the
         second regulated voltage is greater than
         the first regulated voltage;
         providing the first regulated voltage as the
         operating voltage of the memory when the
         first regulated voltage is at least the value
         of the minimum operating voltage; and
         providing the second regulated voltage as
         the operating voltage of the memory when
Case: 22-1906    Document: 72     Page: 5   Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                 5



        the first regulated voltage is less than the
        value of the minimum operating voltage,
        wherein while the second regulated voltage
        is provided as the operating voltage of the
        memory, the first regulated voltage is pro-
        vided to the functional circuit.
 

Id.,

 col. 13, lines 7–28. Independent claim 9 claims a cir-
 cuit, using non-identical but similar language related to
 the points in issue on appeal:
    9. An integrated circuit, comprising:
        a memory that operates using an operating
        voltage, wherein the memory is character-
        ized as having a minimum operating volt-
        age;
        a memory location that stores a value rep-
        resentative of the minimum operating volt-
        age;
        a first voltage regulator for supplying a
        first regulated voltage;
        a circuit that provides a function and uses
        the first regulated voltage;
        a second voltage regulator for supplying a
        second regulated voltage, wherein the sec-
        ond regulated voltage is greater than the
        first regulated voltage; and
        a power supply selector that supplies the
        first regulated voltage as the operating
        voltage of the memory when the first regu-
        lated voltage is at least the minimum oper-
        ating voltage and supplies the second
        regulated voltage as the operating voltage
        when the first regulated voltage is below
        the minimum operating voltage, wherein
        while the second regulated voltage is
Case: 22-1906     Document: 72      Page: 6     Filed: 12/04/2023




 6                 VLSI TECHNOLOGY LLC v. INTEL CORPORATION




         supplied as the operating voltage, the cir-
         cuit uses the first regulated voltage.
 

Id.,

 col. 13, line 59–col. 14, line 15. Those claims are rep-
 resentative for purposes of this appeal.
     The Intel products that are the subject of VLSI’s alle-
 gations of infringement of the ’373 patent are Intel’s
 Haswell and Broadwell microprocessors. Each such micro-
 processor contains a plurality of processor cores that run
 computer programs. It also contains a Ring domain, con-
 taining other circuitry; the Ring domain is sometimes
 called a CLR domain, reflecting that the domain contains,
 though is not limited to, circuitry referred to as CBO cir-
 cuitry, Last Level Cache circuitry, and Ring circuitry. The
 Ring (CLR) domain contains, in addition to the CLR cir-
 cuitry, a static random access memory, i.e., a C6 SRAM.
 The C6 SRAM, as long as it is adequately powered, can
 store information about the state of a core before the core
 goes into low power mode, enabling the core, when it
 “wake[s] back up,” to use the stored information to “pick up
 where [it] left off.” J.A. 1395.
      The accused microprocessors also contain two voltage
 regulators: VCCR and VCCIO. Ordinarily, the VCCR sup-
 plies power to the entire Ring domain, including the C6
 SRAM,         at     variable      power    supply       levels:
 RING_VF_VOLTAGE_0, RING_VF_VOLTAGE_1, and
 RING_VF_VOLTAGE_2. Each of these power supply lev-
 els is stored in a fuse, a type of memory that retains its data
 even if power is removed from it. Another fuse stores a
 different power supply level from the three just mentioned:
 It stores RING_RETENTION_VOLTAGE, which an Intel
 document characterizes as the “worst case retention volt-
 age” for the memory device (the C6 SRAM). J.A. 9574,
 12642; see J.A. 1859, 1951, 2655–57, 2730. If all cores are
 idle, Intel’s accused microprocessors can enter a sleep state
 called Package C7. In that state, substantial evidence
 shows, the cores and CLR components of the Ring domain
Case: 22-1906    Document: 72      Page: 7    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                   7



 are put to sleep, the VCCR voltage is brought below the
 RING_RETENTION_VOLTAGE, and the C6 SRAM re-
 ceives power not from the VCCR but from the VCCIO, at a
 voltage higher than the reduced VCCR voltage. See J.A.
 1388, 1845-49, 1862–63, 1938–40, 2661–66, 3225–26.
                              B
      The ʼ759 patent describes a system in which at least
 two devices, such as computer processors, are coupled to a
 bus that can operate at a variety of frequencies (clock
 speeds); one of the devices, based on its workload (which,
 e.g., indicates a need for faster operations), asks a clock
 controller to change a clock frequency; and the controller is
 programmed to respond by outputting a clock frequency to
 control the speed of the bus and a second device coupled to
 the bus. ʼ759 patent, Abstract. Figure 1 illustrates the
 system:




      In Figure 1, master device 120 is coupled to a bus 102.
 

Id.,

 col. 2 line 66–col. 3, line 3 and col. 3, lines 22–25. A
 programmable clock controller 150 can control the clock
 152 to set the frequency of the bus 102 or devices coupled
 to the bus, including master device 122. 

Id.,

 col. 3, lines
Case: 22-1906     Document: 72      Page: 8    Filed: 12/04/2023




 8                 VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 29–51 and col. 4, lines 20–29. Master device 120 can pro-
 vide a trigger input to the controller, e.g., in response to a
 desired increase in device performance, 

id.,

 col. 3, line 64–
 col. 4 line 8, and the controller can receive the trigger input
 and adjust the clock frequency of the bus or of a second
 master device 122. 

Id.,

 col. 4, lines 42–47.
     Independent claim 14 is representative for present pur-
 poses:
     14. A system comprising:
         a bus capable of operation at a variable
         clock frequency;
         a first master device coupled to the bus, the
         first master device configured to provide a
         request to change a clock frequency of a
         high-speed clock in response to a prede-
         fined change in performance of the first
         master device, wherein the predefined
         change in performance is due to loading of
         the first master device as measured within
         a predefined time interval; and
         a programmable clock controller having an
         embedded computer program therein, the
         computer program including instructions
         to:
             receive the request provided by the
             first master device;
             provide the clock frequency of the
             high-speed clock as an output to
             control a clock frequency of a sec-
             ond master device coupled to the
             bus in response to receiving the re-
             quest provided by the first master
             device; and
Case: 22-1906      Document: 72        Page: 9   Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                      9



              provide the clock frequency of the
              high-speed clock as an output to
              control the variable clock frequency
              of the bus in response to receiving
              the request provided by the first
              master device.
 

Id.,

 col. 8 line 50–col. 9, line 4.
     The Intel products that are the subject of VLSI’s alle-
 gations of infringement of the ’759 patent are various Intel
 microprocessors with “Lake” in their names. The accused
 microprocessors feature cores and a Ring bus connecting
 the cores. As VLSI describes the microprocessors and their
 operation as relevant here, VLSI Br. at 13–15, 35–39, the
 microprocessors also include a power control unit (PCU),
 which is a microcontroller running software called p-code
 (“programs,” VLSI Br. at 13) and which controls the fre-
 quency of the cores and the Ring bus. This control is exer-
 cised when a core monitors its workload and sends a
 Core_Active signal to the power control unit. J.A. 2690–
 92. Based on that signal, core-specific p-code in the power
 control unit (described as one “module” of p-code) calculates
 a speed change, to a higher or lower frequency, and pro-
 vides a request for that speed change to another “module”
 of the p-code software in the power control unit, namely, a
 “decision instructions” module. J.A. 2706–09; see VLSI Br.
 at 14, 38–39. The decision instructions module receives the
 request and in turn outputs a signal from the power control
 unit to change the frequency of the Ring bus or of another
 core on the bus. J.A. 2699–703, 2707–10.
                                 C
     A jury trial was held in late February and early March
 2021. The jury found literal infringement of all the as-
 serted claims of the ’373 patent. J.A. 9. For all the asserted
 claims of the ’759 patent, the jury found no literal infringe-
 ment but found infringement under the doctrine of
Case: 22-1906    Document: 72       Page: 10   Filed: 12/04/2023




 10                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 equivalents. J.A. 9–10. 1 The jury awarded VLSI nonover-
 lapping damages for the infringement of the two patents—
 $1.5 billion for the ʼ373 patent and $675 million for the ʼ759
 patent—each award a lump sum payment for all past and
 future infringement over the life of the patent. J.A. 13–14.
      The district court denied various post-trial motions
 concerning infringement and damages. J.A. 16–64, 74–97.
 The court also ruled, after the trial, on a motion that Intel
 had filed in the fall of 2020, a few months before trial, in
 which Intel (a) sought to amend its answer to assert a de-
 fense that it was licensed to practice both VLSI patents but
 (b) requested that the defense be severed from the rest of
 the case and its adjudication stayed. The motion was based
 on a recent change in ownership of Finjan, Inc., which had
 a license agreement with Intel. Intel argued that the li-
 cense now covered VLSI’s ’373 and ’759 patents because
 VLSI and Finjan were now both under the control of For-
 tress Investment Group LLC. The district court denied the
 motion. J.A. 65–73.
     On April 21, 2022, the district court entered final judg-
 ment. J.A. 98–100. Intel timely appealed. We have juris-
 diction to hear the appeal under 

28 U.S.C. § 1295

(a)(1).
                               II
      We first address the appeals of the infringement ver-
 dicts. We review the verdicts of infringement for substan-
 tial-evidence support.     Lucent Technologies, Inc. v.
 Gateway, Inc., 

580 F.3d 1301

, 1309–10 (Fed. Cir. 2009).
                               A
     With respect to the ʼ373 patent, Intel argues on two
 grounds that the evidence is insufficient to support the


      1The jury also found no willfulness on Intel’s part and
 rejected Intel’s defense of invalidity for anticipation. J.A.
 11–12. Neither ruling is at issue before us.
Case: 22-1906    Document: 72      Page: 11    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                   11



 infringement verdict.       First, Intel argues that the
 RING_RETENTION_VOLTAGE in the accused micropro-
 cessors, which is what VLSI contended is the “minimum
 operating voltage” required by the claims, is not actually
 the minimum voltage at which the C6 SRAM can retain
 data and therefore does not come within the several claim
 limitations requiring a “minimum operating voltage.” Sec-
 ond, Intel argues that the RING_RETENTION_VOLTAGE
 is not used in the microprocessors to determine which volt-
 age to supply to the C6 SRAM (VCCR or VCCIO) and there-
 fore the several “when” limitations (the last two limitations
 of claim 1, the last limitation of claim 9) are not satisfied.
 We reject these arguments.
                               1
      There is ample expert testimony, with adequate sup-
 port in Intel’s internal documents, that Intel’s
 RING_RETENTION_VOLTAGE is the minimum operat-
 ing voltage of the C6 SRAM. At trial, VLSI identified In-
 tel’s C6 SRAM as the claimed “memory” and Intel’s
 RING_RETENTION_VOLTAGE as the claimed “minimum
 operating voltage.” J.A. 2661. VLSI’s expert, Dr. Conte,
 pointed to Intel’s component specifications for the Haswell
 and Broadwell microprocessors, which defined the
 RING_RETENTION_VOLTAGE as the “worst case reten-
 tion voltage” for the Ring domain, and he explained that
 this constituted “the lowest voltage for memory to still re-
 member.” J.A. 2656–57, 9574, 12642. Dr. Conte made ex-
 plicit that the worst case retention voltage is the minimum
 operating voltage for the C6 SRAM, which is in the Ring
 domain. J.A. 2656–57.
     Intel’s argument that this evidence does not constitute
 substantial evidence on the point in dispute is that Intel’s
 C6 SRAM is operational and retains data at a
 RING_VF_VOLTAGE_0 voltage, even when it is lower
 than RING_RETENTION_VOLTAGE. Intel relied criti-
 cally for this argument on a comparison done by its expert,
Case: 22-1906    Document: 72      Page: 12    Filed: 12/04/2023




 12                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 Dr. Sylvester. See, e.g., J.A. 1854–55, 1859–61, 1945–50,
 15342–43. But the jury could reasonably credit VLSI’s ev-
 idence that the comparison presented by Intel was faulty,
 in that the two voltages compared by Dr. Sylvester were
 measured under critically different conditions.
     Specifically, Dr. Conte testified that Dr. Sylvester’s
 comparison         of      RING_VF_VOLTAGE_0               and
 RING_RETENTION_VOLTAGE values does not show that
 RING_VF_VOLTAGE_0               is    ever    lower      than
 RING_RETENTION_VOLTAGE under comparable condi-
 tions, because the relevant memory’s retention voltage is
 (inversely) dependent on temperature, and Dr. Sylvester
 had compared RING_VF_VOLTAGE_0 at 100 degrees Cel-
 sius to RING_RETENTION_VOLTAGE at 0 degrees Cel-
 sius. J.A. 2429–30. Dr. Conte testified: “When you
 compensate for temperature, [RING_VF_VOLTAGE_0] is
 going        to        be        always      above         the
 RING_RETENTION_VOLTAGE.” J.A. 2430. Addition-
 ally, Dr. Conte relied on an Intel technical manual contain-
 ing graphs showing a “Vretention” line and (among other
 things) a “v/f 0” value, the former lower than the latter. See
 J.A. 19243. Dr. Conte testified that Vretention corre-
 sponded to RING_RETENTION_VOLTAGE and v/f 0 cor-
 responded to RING_VF_VOLTAGE_0. J.A. 2426. The jury
 reasonably could credit this evidence and reject Intel’s con-
 tention on the point.
     Accordingly, substantial evidence supports the jury’s
 verdict in favor of VLSI on this point.
                               2
     Intel’s second argument against the sufficiency of the
 infringement evidence regarding the ’373 patent also fails.
 Intel’s argument turns entirely on its contention that the
 claims require that falling below the minimum operating
 voltage be the causal trigger for switching from one voltage
 source to a different one. But that is an argument for a
 claim construction, and Intel sought no claim construction
Case: 22-1906    Document: 72     Page: 13   Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                 13



 on this point. When a claim phrase is not construed, we
 defer to the jury’s view of the claim element unless that
 view is contrary to the only reasonable view of the claim
 element. Avid Technology, Inc. v. Harmonic, Inc., 

812 F.3d 1040

, 1048–49 (Fed. Cir. 2016); Hewlett-Packard Co. v.
 Mustek Systems, Inc., 

340 F.3d 1314

, 1320–21 (Fed. Cir.
 2003). And under that standard, Intel cannot succeed on
 appeal.
     The “when” claim language can reasonably be under-
 stood to mean simply “at the time that.” Dr. Conte testi-
 fied, with support from the Intel “Vretention” document,
 that VCCR is the power source for the C6 SRAM at the time
 that          VCCR           voltage        is         above
 RING_RETENTION_VOLTAGE and that “[w]hen VCCR
 is powered down, the [multiplexer] will switch the [C6
 SRAM] memory over to the VCCIO supply which never
 powers down,” and thereby ensures that the C6 SRAM has
 sufficient voltage to retain data. J.A. 2666; see J.A. 2430–
 31, 2757–58, 2664–70. This was sufficient evidence for the
 jury to find that the voltage source switches at the time
 that VCCR voltage drops below the minimum for C6
 SRAM’s data retention—RING_RETENTION_VOLTAGE.
 Intel did not present evidence requiring a contrary finding
 on those points. And no further explanation of the decision
 mechanism is required in the absence of the claim con-
 struction that Intel now effectively urges but did not seek
 in the district court.
     Accordingly, substantial evidence supports the jury’s
 verdict on this point—and, hence, on infringement of the
 ’373 patent. The judgment of infringement of the ’373 pa-
 tent is therefore affirmed.
                              B
     With respect to the ʼ759 patent, Intel argues on two
 grounds that the verdict of infringement—under the doc-
 trine of equivalents—must be reversed. First, Intel argues
 that prosecution history estoppel bars VLSI’s theory of
Case: 22-1906    Document: 72      Page: 14    Filed: 12/04/2023




 14                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 equivalents. Second, Intel argues that VLSI’s evidence of
 equivalents was legally insufficient. We agree with the sec-
 ond argument and do not reach the first.
                               1
      The doctrine of equivalents provides a limited excep-
 tion to the principle that claim meaning defines the scope
 of the exclusivity right in our patent system: “Applied more
 broadly, the doctrine [of equivalents] would conflict with
 the primacy of the claims in defining the scope of a pa-
 tentee’s exclusive rights.” Sage Products, Inc. v. Devon In-
 dustries, Inc., 

126 F.3d 1420, 1424

 (Fed. Cir. 1997); see
 Warner-Jenkinson Co. v. Hilton Davis Chemical Co., 

520 U.S. 17, 29

 (1997); Aro Manufacturing Co. v. Convertible
 Top Replacement Co., 

365 U.S. 336, 339

 (1961); Johnson &
 Johnston Associates Inc. v. R.E. Service Co., 

285 F.3d 1046, 1052

 (Fed. Cir. 2002) (en banc); Deere & Co. v. Bush Hog,
 LLC, 

703 F.3d 1349, 1356

 (Fed. Cir. 2012). The limits re-
 flect a familiar balance among the importance of preserv-
 ing the public’s ability to rely on claims’ meaning to define
 patent scope, the ability of patentees to protect their inven-
 tions through their claim drafting, and (yet) the occasional
 need to recognize some non-literal scope of protection to
 avoid undermining the exclusivity rights authorized by
 Congress to incentivize certain innovations. See Mahn v.
 Harwood, 

112 U.S. 354, 361

 (1884) (“The public is notified
 and informed . . . that [the patentee’s] claim to invention is
 for such and such an element or combination, and for noth-
 ing more.”); Festo Corp. v. Shoketsu Kinzoku Kogyo Ka-
 bushiki Co., 

535 U.S. 722, 731

 (2002) (“If patents were
 always interpreted by their literal terms, their value would
 be greatly diminished. Unimportant and insubstantial
 substitutes . . . could defeat the patent[.]”); see also Mark-
 man v. Westview Instruments, Inc., 

517 U.S. 370

, 373–74
 (1996); Graver Tank & Mfg. Co. v. Linde Air Products Co.,
 

339 U.S. 605, 607

 (1950). We have explained that liability
 under the doctrine is “exceptional,” Honeywell Interna-
 tional, Inc. v. Hamilton Sundstrand Corp., 

523 F.3d 1304

,
Case: 22-1906    Document: 72     Page: 15    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                  15



 1313 (Fed. Cir. 2008), and “[w]e have emphasized . . . that
 the doctrine of equivalents is the exception, however, not
 the rule,” Eli Lilly & Co. v. Hospira, Inc., 

933 F.3d 1320, 1330

 (Fed. Cir. 2019) (internal quotation marks omitted).
      The exceptional character of the doctrine’s use is main-
 tained by closely related demands that restrict the availa-
 bility of liability under the doctrine. Among them are the
 following. First, proof of equivalents must be limitation
 specific, not focused only on the claim as a whole, though
 the limitation-specific inquiry of equivalence may be in-
 formed by the “role played by each element in the context
 of the specific patent claim.” Warner-Jenkinson, 

520 U.S. at 40

; see DePuy Spine, Inc. v. Medtronic Sofamor Danek,
 Inc., 

469 F.3d 1005, 1017

 (Fed. Cir. 2006); Ethicon Endo-
 Surgery, Inc. v. U.S. Surgical Corp., 

149 F.3d 1309

, 1315
 (Fed. Cir. 1998); Dawn Equipment Co. v. Kentucky Farms
 Inc., 

140 F.3d 1009, 1015

 (Fed. Cir. 1998). Second, for the
 determination of whether a substitute element is only in-
 substantially different from a claimed element and hence
 an equivalent, a traditional formulation—appropriate for
 this case, as VLSI’s use of it indicates—asks “whether a
 substitute element matches the function, way, and result
 of the claimed element.” Warner-Jenkinson, 

520 U.S. at 40

.
 Such matching requires that each of function, way, and re-
 sult be “substantially the same,” see Spectrum Pharmaceu-
 ticals, Inc. v. Sandoz Inc., 

802 F.3d 1326, 1337

 (Fed. Cir.
 2015), with the “way” requirement of particular im-
 portance, as a practical matter, in keeping the doctrine
 properly limited. See Warner-Jenkinson, 

520 U.S. at 35, 39

; Union Paper-Bag Machine Co. v. Murphy, 

97 U.S. 120, 125

 (1877) (stressing the crucial importance of “way”); Ad-
 vanced Steel Recovery, LLC v. X-Body Equipment, Inc., 

808 F.3d 1313, 1320

 (Fed. Cir. 2015) (similar); Zygo Corp. v.
 Wyko Corp., 

79 F.3d 1563, 1569

 (Fed. Cir. 1996) (similar);
 Slimfold Manufacturing Co. v. Kinkead Industries, Inc.,
 

932 F.2d 1453

, 1457–58 (Fed. Cir. 1991) (similar). Third,
 we have long demanded specificity and completeness of
Case: 22-1906    Document: 72     Page: 16    Filed: 12/04/2023




 16               VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 proof as crucial to enforcing the limits on the doctrine: The
 patentee must provide “particularized testimony and link-
 ing argument as to the insubstantiality of the differences
 between the claimed invention and the accused device.”
 Akzo Nobel Coatings, Inc. v. Dow Chemical Co., 

811 F.3d 1334, 1342

 (Fed. Cir. 2016) (internal quotation marks omit-
 ted); see Gemalto S.A. v. HTC Corp., 

754 F.3d 1364, 1374

 (Fed. Cir. 2014); Texas Instruments Inc. v. Cypress Semi-
 conductor Corp., 

90 F.3d 1558

, 1566–67 (Fed. Cir. 1996);
 Lear Siegler, Inc. v. Sealy Mattress Co. of Michigan, 

873 F.2d 1422, 1426

 (Fed. Cir. 1989).
                              2
     VLSI’s proof of equivalence for the ʼ759 patent, though
 limitation specific, was insufficient under those principles.
      VLSI’s equivalents contention is best understood in
 light of its literal-infringement case (which the jury re-
 jected). VLSI’s expert, Dr. Conte, testified that a specified
 core in the accused microprocessors is the “first master de-
 vice” under the claim, the Core_Active signal is the re-
 quired “request to change a clock frequency,” and the power
 control unit is the “programmable clock controller”—to
 which the core “provide[s]” the request (the Core_Active
 signal) and which “receive[s]” that request (the Core_Ac-
 tive signal). J.A. 2694–99. That mapping of claim terms
 onto the accused microprocessors embodied the ordinary
 notion of signals (carrying messages) going from one phys-
 ical component to another. But there was evidence that the
 signal sent by the specified cores is not a request to change
 frequency based on changes the cores identified in their
 own performance, because it is only software running on
 the (receiving) power control unit that, based on observa-
 tions of system conditions, calls for a frequency change. See
 J.A. 2083–85, 2188–89, 2748–51. And the jury eventually
 found no literal infringement. J.A. 9.
     VLSI’s alternative infringement theory, invoking the
 doctrine of equivalents, sought to accommodate the
Case: 22-1906    Document: 72      Page: 17    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                    17



 evidence of the role played by the power control unit in
 making the claim-required request. Under this theory, the
 equivalent of the “first master device” as the provider of the
 “request to change a clock frequency” is the combination of
 (a) the core and (b) the core-specific p-code (a software mod-
 ule) residing on the power control unit that makes the fre-
 quency-change request, and the equivalent of the
 “programmable clock controller” is a different software
 module residing on the same power control unit, namely,
 the “decision instructions,” with this software module re-
 ceiving the request sent by the first software module, both
 software modules being within the power control unit. J.A.
 2705–09. The jury found infringement under the doctrine
 of equivalents. J.A. 10.
      Dr. Conte presented his evidence for equivalence using
 the function, way, result framework, see J.A. 2705, 2707,
 and the appropriateness of that framework is not disputed
 here. See, e.g., Brilliant Instruments, Inc. v. GuideTech,
 LLC, 

707 F.3d 1342

, 1346–47 (Fed. Cir. 2013). Using that
 framework, VLSI had to show that the core and certain
 code, which resides on the power control unit, together per-
 form substantially the same function, in substantially the
 same way, to achieve substantially the same result as the
 claimed “first master device” and that certain other in-
 structions on the same power control unit perform substan-
 tially the same function, in substantially the same way, to
 achieve substantially the same result as the claimed “pro-
 grammable clock controller.” And VLSI had to present par-
 ticularized testimony and linking argument to make that
 showing.
     On the crucial point, Dr. Conte testified as follows:
     MR. HEINRICH (ATTORNEY FOR VLSI). Okay.
     So now, let’s go back to your Doctrine of Equiva-
     lents analysis. Did you apply the function/[way]/re-
     sult test?
Case: 22-1906     Document: 72     Page: 18    Filed: 12/04/2023




 18                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




      DR. CONTE. I did. So let’s put a box around the
      core and the Core 1’s [p]-code. And then this pro-
      vides the same function as required by the claim,
      that is to provide a request.
      Q. And does it provide a request in substantially
      the same way as the claim?
      A. Yeah. The claims says[,] “the first master de-
      vice provides a request.” Now, it’s the first master
      device and its [p]-code that provides the request.
      Q. And how do you characterize the difference be-
      tween those two?
      A. It’s just a difference of where an engineer draws
      this data line.
      Q. Is it a design choice?
      A. It’s a design choice.
      Q. And what’s the result?
      A. Well, the result is that a request is provided.
      Q. And is it the same result in each case?
      A. Yes. No, not really. So[,] each will provide a
      different—you mean in terms of—
      Q. That was a bad question. So[,] is the result of
      the core and its [p]-code sending a request to the
      [power control unit] the same result as what’s re-
      quired by the claim?
      A. Oh, I see. I misunderstood your question. Yes,
      it’s the same result as required by the claim.
 J.A. 2707–08.
     That testimony is insufficient. It contains no meaning-
 ful explanation of why the way in which the request is
 made is substantially the same as what the claim pre-
 scribes. The question is not whether, in a schematic
Case: 22-1906    Document: 72     Page: 19   Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                 19



 drawing used to illustrate functions, an engineer could
 “draw[] . . . [a] line” in different places. The question is
 about actual functionality-location differences. It is not
 enough, moreover, to say that the different functionality-
 location placements were a “design choice.” That label does
 not indicate whether, or begin to explain why, the options
 in the choice are substantially different or substantially
 the same: In both circumstances, the choice between the
 options is a design choice. The question that must be ad-
 dressed is whether the difference in the way the function-
 alities are actually allocated between devices is an
 insubstantial one.
     Here, the claim requires that the request function be
 performed by one component (master device) and the re-
 ceipt and output functions be performed by a distinct com-
 ponent (programmable clock controller). According to
 VLSI itself, what occurs in Intel’s accused microprocessors
 is that the request function is split between two physical
 components (core and power control unit), with the request
 complete only in the second component, and receipt and
 output occur within the second component as well. The re-
 quest provision and the receipt/output are performed not
 by distinct physical components but by different software
 “modules”—one p-code module or a second instruction mod-
 ule. VLSI had to prove—with particularized testimony and
 linking argument—that the elements of the Intel arrange-
 ment were substantially the same as the elements of the
 claimed arrangement. But VLSI offered no meaningful tes-
 timony doing so. The above testimony says nothing re-
 motely sufficient, especially in light of Intel’s evidence
 about the significance of using the power control unit, ra-
 ther than other particular components such as cores within
 the system, for making frequency decisions based on work-
 load information. See, e.g., J.A. 2146–47, 2195–96.
      Based on the evidence presented, VLSI’s doctrine of
 equivalents theory fails as a matter of law. The judgment
 of infringement of the ’759 patent is therefore reversed.
Case: 22-1906    Document: 72       Page: 20   Filed: 12/04/2023




 20               VLSI TECHNOLOGY LLC v. INTEL CORPORATION




                              III
      Intel challenges the award of damages for infringement
 of the ’373 patent. 2 In a pretrial motion under Federal Rule
 of Evidence 702, Intel challenged various aspects of the
 damages analysis set forth by VLSI’s damages expert, and
 it made similar arguments in seeking a new trial after the
 jury verdict. See J.A. 1–2, 3533, 4107–09. As relevant
 here, the district court’s denial of both motions is reviewa-
 ble for abuse of discretion. General Electric Co. v. Joiner,
 

522 U.S. 136, 141

 (1997); Finjan, Inc. v. Secure Computing
 Corp., 

626 F.3d 1197, 1202

 (Fed. Cir. 2010); Fornesa v.
 Fifth Third Mortgage Co., 

897 F.3d 624, 627

 (5th Cir.
 2018). An abuse of discretion exists, for purposes of this
 appeal, if the damages analysis departed from an econom-
 ically sound methodology under the legal principles gov-
 erning royalty damages, overall and as applied, and if that
 departure cannot be deemed harmless. See Summit 6, LLC
 v. Samsung Electronics Co., 

802 F.3d 1283

, 1295–96 (Fed.
 Cir. 2015); VirnetX, Inc. v. Cisco Systems, Inc., 

767 F.3d 1308, 1328

 (Fed. Cir. 2014).
                              A
      Certain well-established principles are relevant to In-
 tel’s challenge to the damages award at issue on appeal—
 which is a (lump-sum life-of-patent) reasonable-royalty
 award designed to compensate “for the use made of the in-
 vention by the infringer.” 

35 U.S.C. § 284

. “The ‘value of
 what was taken’—the value of the use of the patented tech-
 nology—measures the royalty.” Aqua Shield v. Inter Pool
 Cover Team, 

774 F.3d 766, 770

 (Fed. Cir. 2014) (quoting
 Dowagiac Manufacturing Co. v. Minnesota Moline Plow



      2Although Intel also challenges the award of damages
 for infringement of the ʼ759 patent, we need not address
 that challenge because we hold that judgment of non-in-
 fringement of that patent is required.
Case: 22-1906    Document: 72     Page: 21    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                  21



 Co., 

235 U.S. 641, 648

 (1915)). Any reasonable royalty
 must seek to measure the value of the patented technol-
 ogy—it must be “apportion[ed]” to that value—by separat-
 ing out and excluding other value in economic products or
 practices. See Finjan, Inc. v. Blue Coat Systems, Inc., 

879 F.3d 1299, 1309

 (Fed. Cir. 2018); see also Garretson v.
 Clark, 

111 U.S. 120, 121

 (1884); Omega Patents, LLC v.
 CalAmp Corp., 

13 F.4th 1361, 1376

 (Fed. Cir. 2021); Power
 Integrations, Inc. v. Fairchild Semiconductor Interna-
 tional, Inc., 

904 F.3d 965, 977

 (Fed. Cir. 2018); Common-
 wealth Scientific & Industrial Research Organization v.
 Cisco Systems, Inc., 

809 F.3d 1295, 1301

 (Fed. Cir. 2015);
 Ericsson, Inc. v. D-Link Systems, Inc., 

773 F.3d 1201, 1226

 (Fed. Cir. 2014).
      An economically sensible, commonly used method for
 determining this market value posits a “hypothetical nego-
 tiation” between the parties (based on certain assumptions,
 including validity of the patent) to “attempt[] to ascertain
 the royalty upon which the parties would have agreed had
 they successfully negotiated an agreement just before in-
 fringement began.” Lucent Technologies, 

580 F.3d at 1324

.
 The analysis must be one that “tries, as best as possible, to
 recreate the ex ante licensing negotiation scenario and to
 describe the resulting agreement,” recognizing that some
 steps in a sound analysis may involve unavoidable “approx-
 imation and uncertainty.” 

Id. at 1325

 (internal quotations
 omitted); see Dowagiac, 

235 U.S. at 647

. Substantively,
 subject to conditions not in dispute here, “the core economic
 question is what the infringer” in the hypothetical negoti-
 ation “would have anticipated the profit-making potential
 of use of the patented technology to be, compared to using
 non-infringing alternatives.” Aqua Shield, 

774 F.3d at 770

 (italicization of “anticipated” removed); see Carnegie
 Mellon University v. Marvell Technology Group, Ltd., 

807 F.3d 1283, 1304

 (Fed. Cir. 2015) (“A key inquiry in the
 analysis is what it would have been worth to the defendant,
 as it saw things at the time, to obtain the authority to use
Case: 22-1906    Document: 72      Page: 22     Filed: 12/04/2023




 22                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 the patented technology, considering the benefits it would
 expect to receive from using the technology and the alter-
 natives it might have pursued.”).
      Also of relevance to the appeal before us is what we
 have said about the use of licenses in a royalty analysis.
 We have recognized that prices paid in actual licenses may
 have a proper role to play in valuing the patented technol-
 ogy at issue in a case, see Apple Inc. v. Wi-LAN Inc., 

25 F.4th 960, 971

 (Fed. Cir. 2022); Prism Technologies LLC v.
 Sprint Spectrum L.P., 

849 F.3d 1360

, 1368–69 (Fed. Cir.
 2017), but the basic evidentiary precondition is that such
 licenses be “sufficient[ly] comparab[le]” to the royalty pro-
 posed by a party for the technology at issue, Apple, 

25 F.4th at 971

 n.5 (citing Elbit Systems Land & C41 Ltd. v. Hughes
 Network Systems, LLC, 

927 F.3d 1292, 1299

 (Fed. Cir.
 2019)). That requirement often precludes use of other li-
 censes that involve (only or even partly) technology other
 than the patented technology at issue in the case at hand,
 e.g., where there is an inadequate basis for soundly extract-
 ing from such licenses information that is truly informative
 about the value of the technology in the case at hand. See,
 e.g., Apple, 25 F.4th at 973–74; LaserDynamics, Inc. v.
 Quanta Computer, Inc., 

694 F.3d 51, 81

 (Fed. Cir. 2012);
 ResQNet.com, Inc. v. Lansa, Inc., 

594 F.3d 860

, 872–73
 (Fed. Cir. 2010).
                               B
      In the present case, Intel, for its part, presented a dam-
 ages theory based on a comparable-license analysis, J.A.
 2776–84, but we have no issue before us about whether In-
 tel’s analysis met the requirement of comparability in
 terms of the basic economic inquiry into value over alter-
 natives, which could differ from technology to technology.
 Rather, what is before us is VLSI’s damages proof, which
 led to a figure close to what the jury awarded. Intel’s prin-
 cipal challenge to that proof is to the analysis developed by
 VLSI’s experts, which, in four steps, sought, without
Case: 22-1906    Document: 72     Page: 23    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                  23



 reliance on licenses, to identify the incremental value over
 non-infringing alternatives added to Intel’s accused prod-
 ucts by use of the asserted patents and the share of that
 value that Intel and VLSI would have agreed Intel would
 pay in a start-of-infringement hypothetical negotiation.
      The first step in VLSI’s royalty calculation, as applied
 to the ʼ373 patent, was to quantify the effect of use of the
 ʼ373 patent technology on the speed of Intel’s Broadwell
 and Haswell microprocessors, derived from quantifying its
 effect on power savings. J.A. 3446. This quantification was
 based on testing done by one of VLSI’s experts, Dr. Annav-
 aram. See J.A. 1525–87. He tested Intel’s Broadwell and
 Haswell microprocessors and determined that the power
 savings attributable to the ʼ373 patent technology was
 5.45%. J.A. 1593, 2681–82. Dr. Conte (the VLSI expert
 whose testimony on equivalents was discussed supra) then
 translated this power savings benefit into a speed benefit,
 finding a one-to-one ratio for Intel’s Broadwell and Haswell
 microprocessors, so that the 5.45% power savings from use
 of ’373 patent technology provided a 5.45% speed benefit.
 J.A. 2682–83.
     Next, VLSI calculated the effect of a speed improve-
 ment on the price Intel could fetch for its products, includ-
 ing in its study a number of Intel products, not only the
 Broadwell and Haswell microprocessors. VLSI’s damages
 expert, Dr. Sullivan, used a regression model to measure
 this effect of a speed improvement on price. J.A. 1609–13.
 Dr. Sullivan found that, as relevant here, each 1% improve-
 ment in speed was associated with a 0.764% increase in
 price for Intel’s products. J.A. 1613–14, 15277–78.
     Then, based on the quantified effect of the ʼ373 patent
 technology on the speed of Intel’s Broadwell and Haswell
 microprocessors and the calculated effect of a speed im-
 provement on price for Intel’s products, VLSI made a cal-
 culation of the incremental revenue attributable to Intel’s
 use of the ’373 patent technology in its Broadwell and
Case: 22-1906    Document: 72     Page: 24   Filed: 12/04/2023




 24               VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 Haswell microprocessors. Specifically, Dr. Sullivan multi-
 plied the 5.45% speed benefit, by the 0.764% price benefit,
 by the known total infringing revenues for Intel’s Haswell
 and Broadwell microprocessors. J.A. 1655–1656, 15290.
 Dr. Sullivan determined the incremental revenue attribut-
 able to use of the ʼ373 patent technology to be
 $2,115,862,744. J.A. 15290.
      Finally, VLSI had to determine how Intel and VLSI
 would have divided up the calculated incremental revenues
 to set a royalty Intel would have paid VLSI for the right to
 use the patent technology. Reflecting the role of profits
 (not revenues per se) in the inquiry, J.A. 1658–59, 3449,
 Dr. Sullivan concluded that there were no incremental
 manufacturing costs incurred by adopting the technology,
 only other small, incremental costs, e.g., costs of making
 sales, J.A. 1659–60, 3450–51. He then inquired into the
 “relative contributions” of Intel and VLSI to the production
 of the incremental revenues (and hence profits). J.A. 1660;
 see J.A. 3452–53. He did so by considering Intel’s “total
 spending,” including sales and marketing, research and de-
 velopment (R&D), and general and administrative (G&A)
 costs, seemingly allocating a proportionate share to the
 products at issue, and on that basis making “a reasonable
 estimate of Intel’s contribution for purposes of the contri-
 bution apportionment.” J.A. 3453, 1661–62; see J.A. 3461.
 The Intel contribution figure was 23.8% of total revenue for
 the two products, with a 76.2% contribution figure for VLSI
 (actually, VLSI’s predecessor in interest, Freescale). J.A.
 1662, 3461–62. Multiplying the VLSI figure by the incre-
 mental revenue gave the reasonably royalty. J.A. 1662–64,
 3461–62, 15301–03. The net result was a proposed royalty
 of $1,611,609,964. J.A 15303. The jury awarded $1.5 bil-
 lion. J.A. 13.
                              C
     Intel challenges several aspects of the foregoing analy-
 sis, but it suffices for vacatur of the award that, in
Case: 22-1906    Document: 72     Page: 25    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                  25



 determining the power savings attributable to use of the
 ʼ373 patent, Dr. Annavaram made a readily identifiable er-
 ror. The step at issue departed from the essential logic of
 the value-of-the-patented-technology assessment. It is not
 a matter of choosing one reasonable step over another or of
 estimation in the face of acceptable imprecision or uncer-
 tainty.
      VLSI’s damages model required VLSI to calculate the
 incremental technical benefit attributable to Intel’s in-
 fringement. For the ʼ373 patent, Dr. Annavaram pur-
 ported to calculate this benefit by calculating the power-
 savings benefit attributable to the accused processor func-
 tion of using the multiplexer to change the source of voltage
 to the C6 SRAM (allowing all cores to go to sleep without
 loss of re-startup data, thereby saving power)—what the
 parties before us call using the C6 SRAM. J.A. 1557–58.
 It is undisputed that when Intel’s accused microprocessor
 enters the Package C7 sleep state, in which all cores are
 asleep, that function is performed (the C6 SRAM is used),
 but that it is not performed when Intel’s accused micropro-
 cessor enters the Core C7 sleep state, in which individual
 cores are asleep. J.A. 1559–60. Thus, Core C7 and Pack-
 age C7 are different states, and only the latter reflects the
 benefits of the infringement at issue. Nevertheless, Dr.
 Annavaram made use of Core C7 state residency data—
 data on how much time a processor spends in a given
 state—in making a choice of inputs into his calculation.
     Specifically, he ran experiments on six Intel devices,
 two with accused Broadwell processors (the other four not
 accused of infringement of the ’373 patent), to collect data
 on the residency of the devices in particular states when he
 put them through selected workloads. He compiled the re-
 sults in a table, one line of which showed residency in the
 Core C7 state, while another line residency in the Package
 C7 state. See J.A. 3132–33. The residency figures for the
 former (where any core is shut down, J.A. 18670–71) were
 close to double those of the latter (where all cores are shut
Case: 22-1906    Document: 72      Page: 26    Filed: 12/04/2023




 26                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 down, J.A. 18671–72). Then, in the next stage of his anal-
 ysis, which was to employ an Intel analytic tool (Power
 Model) that estimates power use (and hence power savings)
 under specified conditions (including residency and work-
 load), Dr. Annavaram used at least the Core C7 figures,
 and perhaps also the Package C7 figures, in carrying out
 some kind of “match[ing]” process in order “to select the
 residency and workload settings in the Intel Power Model.”
 J.A. 1580–81; see J.A. 1533–34, 1555–61, 3137–41. 3
     The results of using the Power Model with the selected
 inputs were the power savings that were crucial to VLSI’s
 damages calculation. Yet to produce the power-savings re-
 sults, Dr. Annavaram used inputs that he chose by trying
 to match (only or in part) data not from use of infringing
 functionality. That step undermines the reliability of the
 results as a calculation of power savings from use of the
 infringing functionality.
     We cannot say that this error “could not have changed
 the result,” namely, the precise amount of damages, so as
 to render it harmless. See Wright & Miller § 2886; Voda v.
 Cordis Corp., 

536 F.3d 1311, 1328

 (Fed. Cir. 2008); Cy-
 toLogix Corp. v. Ventana Medical Systems, Inc., 

424 F.3d 1168, 1174

 (Fed. Cir. 2005); Ecolab, Inc. v. Paraclipse, Inc.,
 

285 F.3d 1362, 1374

 (Fed. Cir. 2002). The difference be-
 tween the Core C7 residency data and the Package C7 res-
 idency data is on its face significant—75.86% and 75.95%
 for the Core C7 state versus 41.01% and 57.49% for the
 Package C7 state. J.A. 3132. Residency in particular
 states matters to the power savings, see J.A. 3138, and res-
 idency inputs chosen in an effort to match non-infringing-


      3Dr. Annavaram stated at least once that the Core C7
 state was “the one” state he used from his experiment-re-
 sults table. J.A. 1579–81. The problem we identify exists
 even if Dr. Annavaram used that state along with the Pack-
 age C7 state (or others).
Case: 22-1906    Document: 72      Page: 27    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                   27



 state numbers (at least in part) could well affect the bottom
 line. See also J.A. 1975–76, 1994 (testimony of Intel’s ex-
 pert, Dr. Sylvester). On this record, we cannot deem this
 step in the damages calculation harmless as to the bottom-
 line amount of damages. The damages award must be set
 aside, without our adjudicating Intel’s other allegations of
 error bearing on the damages.
                               D
      The proper relief is a remand for a new trial on the is-
 sue of damages. We see no sound basis, given the nature
 of the error found or of the other errors alleged, for denying
 VLSI an opportunity to provide a corrected damages case.
 Nor do we see a sound basis, in the error found or the other
 errors alleged, for setting aside the liability verdict as in-
 fected by the damages verdict.
     We do not address Intel’s other allegations of error in
 VLSI’s damages case, except to state the following, which
 bears on what should occur on remand. First, Intel chal-
 lenges VLSI’s introduction of certain concededly noncom-
 parable licenses in the trial, which were not part of the
 experts’ damages calculation described supra. Given the
 conceded noncomparability, the law restricting use of
 noncomparable licenses, see supra p. 22, would clearly bar
 admission of that evidence unless admission of the evi-
 dence could be justified on the ground—which VLSI urges
 here and which we read the district court as having
 adopted—that the evidence was a proper response to Intel’s
 presentation, through cross-examination of VLSI’s expert,
 of prices at which certain sports teams were sold. See J.A.
 2328–29; J.A. 17. The Intel line of questioning was allowed
 over VLSI’s objection and unavoidably conveyed a message
 of relevance of the sheer size of the proposed award,
 through facts (about sports teams) self-evidently uncon-
 nected to the question of the value of the specific technology
 at issue. We do not decide whether VLSI’s evidence of
 noncomparable licenses, itself focusing only on sheer size
Case: 22-1906     Document: 72      Page: 28    Filed: 12/04/2023




 28                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 (in VLSI’s case, of license payments Intel made for other
 technology), could be admitted as a response to Intel’s
 presentation. And we need not do so. We expect that, on
 remand, there will be no recurrence of comparable refer-
 ences, to sports-team prices or other facts about size of a
 proposed award alone (one way or the other), not focused
 on answering the question of the value of the specific tech-
 nology at issue.
      Second, we similarly do not adjudicate the merits of In-
 tel’s other challenges to VLSI’s damages-calculation meth-
 odology. We note only this: Intel, in this court, has not
 persuasively shown that the regression analysis used to de-
 termine price effects of speed improvements is an improper
 or unreasonable one, and VLSI has not adequately eluci-
 dated how the last-step cost-and-contribution analysis (see
 supra, section III.B) reasonably establishes the choice Intel
 and VLSI would have made in the hypothetical negotiation
 about the sharing of the incremental benefits of implemen-
 tation of the patent technology. On remand, the oppor-
 tunity to provide better explanations should be made
 available.
                               IV
      We address, finally, Intel’s license defense.
     On September 4, 2020, as the case was approaching
 trial (which began in late February 2021), Intel filed a mo-
 tion to stay the litigation, arguing that it had newly ac-
 quired a license defense to the alleged infringement. Intel
 alleged that it had a license to practice VLSI’s asserted pa-
 tents because, as relevant, (a) in 2012, Intel entered into
 an agreement with Finjan, in which Finjan granted Intel a
 perpetual and irrevocable license to patents owned and
 controlled by Finjan’s “affiliates” and (b) in July 2020, For-
 tress acquired “control” of Finjan, through funds Fortress
 manages, making VLSI and Finjan “affiliates” under the
 license agreement. J.A. 3005–06; see J.A. 65–66. On No-
 vember 10, 2020, Intel filed the motion now at issue: It
Case: 22-1906    Document: 72      Page: 29    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                   29



 sought leave to amend its answer in this case to add the
 license defense, and it requested that the defense be sev-
 ered from the rest of the case and its adjudication stayed.
 J.A. 3635–36. The district court ruled on the motion only
 after trial, denying it on March 18, 2022. J.A. 65. We now
 reverse that denial.
     The license agreement invoked by Intel was entered
 into on November 20, 2012, between Intel, on one side, and
 Finjan Software, Inc. and Finjan, Inc., on the other, and in
 the license agreement “Finjan” granted to Intel a broad li-
 cense to “Finjan’s Patents.” J.A. 3684–96. The license
 agreement defines “Finjan” as Finjan Software, Inc., Fin-
 jan, Inc., and their “Affiliates”; it defines “Affiliates” as
 “any Person that, now or hereafter, directly or indirectly
 through one or more entities, controls or is controlled by, or
 is under common control with” “Finjan”; and it defines
 “control” to mean “the possession, direct or indirect, of the
 power to direct the management and policies of a Person,
 whether through the ownership of any percentage of voting
 interests of such Person, through contract or otherwise.”
 J.A. 3684. The license agreement also broadly defines “Fin-
 jan’s Patents” as “all Patent Rights that are owned or con-
 trolled at any time on or after November 6, 2012[,] by
 Finjan.” J.A. 3685. Intel alleged that on July 24, 2020,
 Fortress “funds acquired Finjan, Inc., causing Finjan and
 VLSI to be under the common control of Fortress, and
 therefore ‘affiliates’ under the provisions of the license.”
 J.A. 3635.
     Addressing Intel’s motion, the district court applied the
 four-factor approach set forth in Meaux Surface Protection,
 Inc. v. Fogleman, 

607 F.3d 161, 167

 (5th Cir. 2010), consid-
 ering (1) the explanation for the failure to timely move for
 leave to amend; (2) the importance of the amendment; (3)
 potential prejudice in allowing the amendment; and (4) the
 availability of a continuance to cure such prejudice. J.A.
 67–72. The district court denied the motion on the grounds
Case: 22-1906    Document: 72      Page: 30    Filed: 12/04/2023




 30                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 of untimeliness, prejudice, and—most significantly—unim-
 portance because of the futility of the defense. 

Id.

      We follow regional circuit law in reviewing the denial
 of a motion to amend, Healthier Choices Management Corp.
 v. Philip Morris USA, Inc., 

65 F.4th 667, 675

 (Fed. Cir.
 2023), and Fifth Circuit law provides for review here for
 abuse of discretion, Meaux, 

607 F.3d at 167

. A district
 court abuses its discretion if it bases its ruling on an erro-
 neous view of the law or on a clearly erroneous assessment
 of the evidence. Cooter & Gell v. Hartmarx Corp., 

496 U.S. 384, 405

 (1990). We find such an abuse of discretion.
      We conclude that the district court’s conclusion that In-
 tel unduly delayed filing its motion—between the time of
 the July 24 acquisition and the filing of the November 10
 motion—was an abuse of discretion. Intel was required by
 the 2012 license agreement to follow certain procedures,
 J.A. 3694, and Intel acted with diligence in doing so. On
 August 17, 2020, in accordance with the process require-
 ments, Intel sent a letter to Finjan, VLSI, and Fortress
 stating that Intel holds a license to VLSI’s patents under
 the 2012 Intel-Finjan license agreement and that Intel
 would like to begin the dispute resolution process. J.A.
 3018. Then, on September 2, 2020, Intel filed its motion to
 stay in light of the new license defense. J.A. 3001. This
 motion to stay was in accordance with the contract’s choice
 of law and venue provisions. J.A. 3696. VLSI opposed the
 motion on September 18, 2020, see J.A. 156, and when the
 district court had not ruled by early November, Intel filed
 the motion now at issue on November 10, 2020, stating that
 it moved to amend its answer “[t]o avoid any doubt that
 Intel has preserved its defense of license in this case.” J.A.
 3635. Considering timing alone, we see no reasonable ba-
 sis for a determination of undue delay.
     Regarding prejudice, the district court, which expressly
 considered its futility conclusion in addressing this factor,
 did not conclude that, even if the defense was meritorious,
Case: 22-1906    Document: 72      Page: 31     Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                    31



 it would deny the motion because of prejudice to VLSI
 alone. J.A. 71. And we see no basis on which prejudice
 alone could support denial of the motion. Intel requested
 severance of the defense from the rest of the case and a stay
 of its adjudication, so trial of the other issues was not to be
 delayed. Although entry of a final judgment in the district
 court for appeal might have been delayed, the district court
 said nothing about how long such a delay would have been.
 The court said that having a district court judgment sooner
 rather than later (subject to appeal) would help VLSI se-
 cure licenses, but it offered no basis for that conclusion that
 is independent of the merits of the defense. Nor did the
 district court consider that it was Fortress (with its close
 relationship to VLSI), not Intel, that was responsible for
 the action creating the possibility of the license defense:
 “Fortress acquired Finjan,” J.A. 70. The district court con-
 sidered relevant to the prejudice factor Intel’s interest in
 not being liable if its defense is valid, but it dismissed that
 interest only by an unelaborated reference to the possibil-
 ity of a contract action by Intel against Finjan as an alter-
 native remedy for Intel, without explaining why that was
 a meaningful avenue (legally or practically) for recovering
 from Finjan the damages Intel would pay in this case (in-
 correctly if the license defense was valid). For such rea-
 sons, we see no basis on which, futility aside, prejudice
 could sustain the denial of Intel’s motion.
     What is determinative, then, is the soundness of the
 district court’s conclusion that the record developed makes
 the license defense so clearly meritless that allowing it
 even to be pleaded is a futile act. We hold that conclusion
 to be wrong as a matter of law. This is a very narrow hold-
 ing. We do not conclude that the license defense is merito-
 rious. We conclude only that the governing law is such that
 the defense requires additional litigation of the sort that
 begins once it is added to the case, whether that process is
 a fully developed motion to dismiss, with fuller analysis of
 the governing law than has yet occurred, or more fact-
Case: 22-1906    Document: 72      Page: 32    Filed: 12/04/2023




 32                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 based litigation. Intel might well fail to sustain the de-
 fense, but we do not see failure as foreordained on the ma-
 terial supplied to date.
      The district court turned to Delaware law, which, sub-
 ject to any overriding federal law, is the applicable law cho-
 sen by the 2012 license agreement. J.A. 3696. The district
 court stated, without qualifiers, that “Delaware law pro-
 vides that a non-party to a contract is not bound by that
 contract.” J.A. 70. But the precedential authority cited in
 support states the principle as only “the ordinary rule,”
 subject to exceptions for “unusual circumstances.” Alliance
 Data Systems Corp. v. Blackstone Capital Partners V L.P.,
 

963 A.2d 746

, 760 & n.47 (Del. Ch. 2009), affʼd, 

976 A.2d 170

 (Del. 2009) (footnote quoting statement from Wallace
 ex rel. Cencom Cable Income Partners II, Inc., v. Wood, 

752 A.2d 1175, 1180

 (Del. Ch. 1999), that this is “a general prin-
 ciple”). Three cited nonprecedential decisions are to the
 same effect. Sheehan v. AssuredPartners, Inc., No. 2019-
 0333-AML, 

2020 WL 2838575

, at *9 (Del. Ch. May 29,
 2020) (“‘ordinary rule,’” quoting Alliance Data, 963 A.2d at
 760–61); Wenske v. Blue Bell Creameries, Inc., No. 2017-
 0699-JRS, 

2018 WL 5994971

, at *3 (Del. Ch. Nov. 13, 2018)
 (“ordinarily”); Kuroda v. SPJS Holdings, L.L.C., No. 4030-
 CC, 

2010 WL 4880659

, at *3 (Del. Ch. Nov. 30, 2010)
 (“[g]enerally”).
      In 2019, the Delaware Supreme Court stated that
 “[c]ontracts may impose obligations on affiliates” in certain
 contexts, and it affirmed the Delaware Chancery Court’s
 determination that the case before it involved such a con-
 text. In re Shorenstein Hays-Nederlander Theatres LLC
 Appeals, 

213 A.3d 39, 57

 (Del. 2019). In so ruling, the Del-
 aware Supreme Court cited, 

id.

 at 57 n.86, two nonprece-
 dential decisions of the Delaware Chancery Court that are
 to the same effect, describing them as involving contract
 provisions covering certain affiliates and non-signatories
 who were, or came to be, owned or under the control of a
 signatory party.        Medicalgorithmics S.A. v. AMI
Case: 22-1906    Document: 72     Page: 33    Filed: 12/04/2023




 VLSI TECHNOLOGY LLC v. INTEL CORPORATION                  33



 Monitoring, Inc., No. 10948-CB, 

2016 WL 4401038

, at *18
 (Del. Ch. Aug. 18, 2016); MicroStrategy Inc. v. Acacia Re-
 search Corp., No. 5735-VCP, 

2010 WL 5550455

, at *12 (Del.
 Ch. Dec. 30, 2010).
     This case law does not definitively enough answer
 questions of potential significance here to make the license
 defense proposed by Intel futile. Perhaps it makes a differ-
 ence whether the affiliate sought to be bound was an affil-
 iate at the time of contract or whether it controlled or was
 controlled by the signatory, rather than merely under com-
 mon control, even when the contractual definition of “affil-
 iate” includes the common-control situation. Under the
 authorities presented and arguments made on whether
 VLSI, as a non-party, could be bound by the 2012 license
 agreement, we do not think that there is a sufficiently clear
 answer for there to be a determination of futility. See
 Stripling v. Jordan Production Co., LLC, 

234 F.3d 863

,
 872–73 (5th Cir. 2000).
     In so concluding, we do not prejudge the answer to that
 question or the answer to other questions about whether
 VLSI’s particular circumstances bring it within the con-
 tract’s terms. Nor do we prejudge the appropriate pro-
 cesses for deciding the merits of the license defense, e.g.,
 whether a motion to dismiss may suffice, whether discov-
 ery is warranted, or whether summary judgment may be
 justified. Nor do we prejudge whether certification of a le-
 gal question to the Delaware Supreme Court is appropriate
 or whether any federal patent-law question is involved.
 See, e.g., 

35 U.S.C. § 261

. We hold only that it was error to
 deny the motion to add the license defense to the case.
                              V
     The judgment of infringement of the ʼ373 patent is af-
 firmed. The judgment of infringement of the ʼ759 patent is
 reversed. The damages award for the ʼ373 patent is va-
 cated. The denial of the motion for leave to amend is
Case: 22-1906     Document: 72    Page: 34      Filed: 12/04/2023




 34                VLSI TECHNOLOGY LLC v. INTEL CORPORATION




 reversed. The matter is remanded for further proceedings
 consistent with this opinion.
      The parties shall bear their own costs.
      AFFIRMED IN PART, REVERSED IN PART,
        VACATED IN PART, AND REMANDED