Wisconsin Alumni Research Foundation v. Apple Inc.

Filing 193

ORDER: Defendant and counter claimant Apple, Inc.'s motion for summary judgment (dkt. # 116 ) is denied as to its counterclaims and defenses of anticipation by Steely and indefiniteness, and denied as to plaintiff's willful infringement cla im premised on (1) Apple's claim construction, (2) anticipation by Steely, and (3) indefiniteness of claims 5 and 6. The court reserves on the motion in all other respects.Plaintiff's motion for summary judgment (dkt. # 117 ) is granted. Signed by District Judge William M. Conley on 8/5/2016. (voc)

IN THE UNITED STATES DISTRICT COURT
FOR THE WESTERN DISTRICT OF WISCONSIN
WISCONSIN ALUMNI RESEARCH
FOUNDATION,
Plaintiff,

OPINION AND ORDER

v.
14-cv-062-wmc

APPLE, INC.,
Defendant.

In this lawsuit, plaintiff Wisconsin Alumni Research Foundation (“WARF”)
alleges that defendant Apple, Inc. infringes U.S. Patent No. 5,781,752 (the “’752
patent”), which concerns a “table based data speculation circuit for parallel processing
computer.” Before the court are the parties’ cross-motions for summary judgment and
claim construction. (Dkt. ##116, 117.) For the reasons that follow, the court will adopt
WARF’s proposed construction of the term “prediction” and grant summary judgment to
WARF on Apple’s counterclaims and defenses based on anticipation under 35 U.S.C.
§ 102 with respect to U.S. Patent No. 5,619,662 (“Steely” or the “Steely patent”), as
well as indefiniteness under 35 U.S.C. § 112 ¶ 2 with respect to claims 5 and 6. In turn,
the court will deny Apple’s motion for summary judgment based on those same defenses
and counterclaims. As for Apple’s motion for summary judgment on WARF’s claim of
willful infringement, the court will deny Apple’s motion with respect to any defenses
premised on (1) Apple’s claim construction, (2) anticipation by Steely, and (3)
indefiniteness of claims 5 and 6, but will reserve on Apple’s motion in all other respects.

UNDISPUTED FACTS1
A. The Parties and Overview of This Lawsuit
Plaintiff Wisconsin Alumni Research Foundation (“WARF”) is a Wisconsin
corporation, with its principal place of business in Madison, Wisconsin.

WARF is the

owner of the ’752 patent. Defendant Apple, Inc. is a California corporation, with its
principal place of business in Cupertino, California.
On January 31, 2014, WARF filed suit against Apple alleging infringement of the
’752 patent. Apple answered and asserted counterclaims for declaratory judgment of
non-infringement and invalidity of the ’752 patent. Material to the present motions,
Apple contends that claims 1-3, 5, 6, and 9 of the ’752 patent are invalid as anticipated
by the “Steely patent. Apple also alleges that claims 5 and 6 of the ’752 patent are
invalid as indefinite.
B. Technology Overview
A modern computer device includes both hardware and software.

Hardware

typically includes memory, a microprocessor and peripherals, while software typically
consists of sequences of instructions or “programs” that run on the hardware.

At a

general level, the microprocessor is responsible for fetching instructions and data,
executing those instructions to modify the data, and then saving the results.2 Typically,

1

Except as otherwise noted, for purposes of summary judgment, the court finds the following
facts to be material and undisputed.
2

While the court uses the term “executing,” the court acknowledges that the parties have agreed
on a construction for the term “in fact executed” described below in subheading “E” of this Facts
section.

2

individual instructions call for the performance of a relatively simple task, such as reading
a value from or writing a value to a memory location, or adding, subtracting or comparing
two numbers.

There are generally three types of software instructions: (1) memory

instructions; (2) computing instructions; and (3) control instructions.
Memory instructions are instructions that “when executed, cause data to be
loaded into the processing unit from memory or stored from the processing unit to
memory.” (’752 Patent (dkt. #1-1) 1:38-40.) So-called “LOAD” instructions copy or
read a value stored at a memory location specified by an address and return a value.
LOAD instructions are also called “data consuming instructions,” because they consume
data by obtaining data from memory, though as Apple cautions, other types of software
instructions also “consume” data. “STORE” instructions, on the other hand, copy or
write a value to a memory location specified by an address. For that reason, STORE
instructions are also called “data producing instructions,” as they produce data by
providing data to memory. (Apple similarly points out that other types of instructions
“produce” data.) When a STORE instruction executes, it overwrites any value previously
stored at that memory location.

Both LOAD and STORE instructions are memory

instructions.
Generally speaking, software instructions in a program have a predefined “program
order,” where the processor performs the instructions sequentially.

Instructions,

however, need not always be executed in the listed order. Instead, they may be executed
“out of order.” In out-of-order executions, instructions are typically executed when ready
-- in other words, based on the availability of their input data, or “operands,” rather than
3

a specified program order.3 There are some obvious benefits to permitting instructions to
execute out of order. For instance, because some instructions in a program take longer to
execute than others, performing instructions in program order may slow processor
performance since it requires waiting for earlier instructions to execute before performing
later instructions in the program.

Out-of-order execution, therefore, may result in

increased efficiency since it allows the processor to use free time to execute other
instructions that are ready to be processed. On the other hand, out-of-order execution
may have a detrimental effect on performance if it leads to errors that require the
processor to expend resources to correct.
A key requirement of efficient out-of-order execution, therefore, is that it must
yield the same results as would the execution of instructions in program order. This
requirement touches on the concept of “instruction dependency.”

A dependent

instruction is one that must wait for the result of an earlier-in-order execution before it
can safely execute.4

For example, data dependency exists when an earlier-in-order

STORE instruction writes data to the same address that is accessed by a later-in-order
LOAD instruction. In that situation, the STORE and LOAD must execute in program
order for the LOAD to read the correct data from the shared memory address that both
instructions access.
3

Apple clarifies that out-of-order executions also depend on the availability of necessary
hardware.
4

While “[a] processor may permit dependent instructions to execute out-of-order and then
invoke a recovery process to return to a correct machine state,” as Apple describes, Apple fails to
dispute WARF’s point that to execute “safely,” the dependent instruction must wait to execute
until after the instruction on which it depends has been executed. (Pl.’s Reply to Pl.’s PFOFs
(dkt. #157) ¶ 21.)

4

In some situations, whether a given LOAD instruction depends on a STORE
instruction from an earlier earlier-in-order program step cannot be known until after one
or both of the instructions are executed. In other words, the processor lacks sufficient
information to resolve whether or not a dependency actually exists. This uncertainty is
known as “ambiguous dependency.” Ambiguous dependencies may occur, for example,
when the memory addresses that must be accessed by a given LOAD or STORE
instruction are computed “on the fly” as the program executes. In those circumstances,
the processor may have to perform additional computations with data that are not
currently available in order to resolve whether one instruction is dependent on another.
To maximize processing speed, however, the processor may elect to execute a
LOAD instruction before an earlier STORE instruction. The out-of-order execution of
instructions without knowing if there is an actual dependency between them is known as
“speculation” or “speculative execution,” because the processor is speculating that there is
no actual dependency. Speculation can be advantageous if it turns out to be correct (i.e.,
the LOAD instruction in fact was not dependent on the STORE instruction); then the
out-of-order execution will yield the correct result and the performance will improve. 5 In
contrast, if a LOAD instruction is speculatively executed ahead of a STORE instruction
of earlier program order and it turns out that the speculation was incorrect (i.e., the
5

The parties dispute whether “it is quite often the case that an ambiguous dependency is resolved
as no dependency at all,” as the ’752 patent represents. (’752 patent (dkt. #1-1) 2:26-27.) As
Apple contends, “[t]he degree to which ambiguous dependencies will turn out to be resolved as no
dependency depends upon the workload.” (Def.’s Resp. to Pl.’s PFOFs (dkt. #141) ¶ 30.) At the
same time, Apple also proposes on summary judgment that “[m]any program instructions are
‘independent’ of each other and can safely execute-out-of-order with respect to each other.”
(Def.’s PFOFs (dkt. #119) ¶ 24.) Even if this could be construed as a dispute, it is not material
to the issues before the court on summary judgment.

5

LOAD instruction was in fact dependent on the earlier STORE instruction), then the
instructions will cause an error -- the prematurely executed LOAD instruction having
obtained incorrect or stale data.6
In the patent-in-suit, this error is referred to as “mis-speculation,” although the
Steely patent -- as described below -- refers to it as a “collision.” As discussed generally
already, and as the patent-in-suit explains specifically, a mis-speculation can be
detrimental to processor performance because it requires “the results of the prematurely
executed dependent instructions [to] be discarded” or “squashed,” and the instruction
will need to be re-executed in program order. (’752 patent (dkt. #1-1) 2:46-49; Def.’s
PFOFs (dkt. #119) ¶ 33.)7
C. The ’752 Patent
i.

Overview and Prosecution History

The ’752 patent, entitled “Table Based Data Speculation Circuit for Parallel
Processing Computer,” was filed on December 26, 1996, and issued on July 14, 1998.
The listed inventors are Drs. Andreas I. Moshovos, Scott E. Breach, Terani N.
Vijaykumar, Gurindar S. Sohi. Plaintiff WARF is listed as the original assignee. WARF
maintains that the named inventors conceived of the claimed invention no later than
December 11, 1995.

6

Apple maintains that there will be no error, at least technically, if the yet-to-execute STORE
instruction would not change the value already written to the memory address, because the
LOAD instruction still obtains the correct data. (Def.’s Resp. to Pl.’s PFOFs (dkt. #141) ¶ 31.)
7

Apple contends that there may be times that the performance cost of mis-speculation does not
outweigh the performance benefit of speculation. (Def.’s Resp. to Pl.’s PFOFs (dkt. #141) ¶ 34.)

6

During prosecution of the ’752 patent, the named inventors provided no prior art
to the Patent Office.

On October 8, 1997, the patent examiner issued a Notice of

References Cited, which listed four pieces of prior art. The patent examiner rejected
pending claims 1-2, 6 and 8-11 as anticipated in light of U.S. Patent No. 5,555,432
(“Hinton”). On January 5, 1998, WARF filed a response cancelling pending claims 9 and
10, but arguing that claims 1, 2, 6, 8 and 11 were allowable over Hinton. On February 3,
1998, the examiner allowed those claims.
ii.

Objectives and Specification

The ’752 patent recognizes that “[t]he performance cost is a function of the
frequency that speculation is required, the probability of mis-speculation and the time
required to recover from a mis-speculation.” (’752 patent at 3:14-18.) The ’752 patent
also observes that “most data dependent mis-speculations can be attributed to a few
static STORE/LOAD instruction pairs,” and that mis-speculations typically “exhibit
‘temporal locality,’” such that “if one LOAD/STORE pair causes a data mis-speculation at
a given point in time, it is highly likely that a later instance of the same pair will soon
cause another mis-speculation.” (Id. at 3:51-57.) The patent further observes that
The present inventors believe that a relatively limited number
of LOAD/STORE pairs will create mis-speculation and so the
operation described above prevents the majority of the
LOAD/STORE pairs from being slowed in execution. The list
of critical LOAD/STORE pairs is prepared dynamically in a
synchronization method for those LOAD/STORE pairs . . . .

7

(Id. at 14:15-22.) Based on these observations, the inventors concluded that load-based
memory dependencies may be amenable to history-based prediction.8 As such, the ’752
patent associates predictions with particular LOAD instructions that have mis-speculated
in the past.
The specification of the ’752 patent describes a processor containing a “data
speculation circuit” that detects dependence between LOAD and STORE instructions.
The data speculation circuit also detects mis-speculations where a LOAD instruction that
is dependent for its data on a STORE instruction appearing earlier in program order is in
fact executed before the STORE instruction. According to the preferred embodiment,
the data speculation circuit sends a “mis-speculation indication” to a “predictor circuit” if
a mis-speculation is detected, which uses the indication to then produce a prediction.
The greater the “prediction,” the greater the likelihood that the speculative execution of
its associated LOAD instruction will cause a mis-speculation; the lower a prediction at a
given time, the lower the likelihood of mis-speculation.

The processor uses each

prediction to decide whether its associated LOAD instruction should be allowed to
execute speculatively.
The patent discloses a “three-tiered approach” to dealing with ambiguous
dependency. The first tier considers whether a LOAD instruction has a history of misspeculation. “If there is no history of data mis-speculation, [the instruction] is executed
without further inquiry.” (’752 patent (dkt. #1-1) 3:64-66.) At this tier, the ’752 patent

8

Apple disputes that the named inventors were the first to develop history-based techniques for
load-based memory dependencies. (Def.’s Resp. to Pl.’s PFOFs (dkt. #141) ¶ 42.)

8

describes a “prediction table,” in which entries are created when the processor detects a
mis-speculation by a LOAD instruction. “[I]f no entry is found in the prediction table,”
then “the reasonable assumption is that speculation can proceed.” (Id. at 11:22-24.)
The second tier becomes relevant with a LOAD instruction has mis-speculated in the
past. In this tier, “a predictor based on the past history of mis-speculations for that
LOAD instruction is employed to determine whether the instruction should be executed
or delayed.” (Id. at 4:1-4.) With respect to the second tier, the patent explains that “it is
an object of the invention to provide a predictor circuit that may identify data
dependencies on an on-going and dynamic basis.” (Id. at 4:31-33.) Finally, in cases
where the prediction indicates that the LOAD instruction should not be executed
speculatively, the third tier may be employed to decide when the LOAD instructions
should be allowed to execute. This part of the patent describes a “synchronization table,”
which “indicates whether there is in fact a pending LOAD instruction awaiting its
dependent STORE instruction.” (Id. at 11:45-47.)
iii.

Claim Construction
a) “Prediction”

The heart of the parties’ dispute turns on the meaning of the term “prediction” as
used in claim 1 and all other independent claims.

As context, WARF contends that the

term “prediction” should be construed to mean “a variable that indicates the likelihood
that the data speculative execution of a load instruction will result in a mis-speculation,”
where “a ‘prediction’ must be capable of receiving ongoing updates.” (Pl.’s PFOFs (dkt.
#122) ¶ 67.)

In contrast, Apple contends that “prediction” need not be capable of
9

receiving updates, and it, therefore, proposes a construction of “a value that indicates
that likelihood that the data speculative execution of a load instruction will result in a
mis-speculation,” but does not necessarily contemplate a revision to that value based on
regular updates. (Id. at ¶ 68.)
b) Other Agreed-Upon Terms
The parties agree to the following constructions of claim terms:


“data speculation circuit” (claims 1 and 9): “a circuit that detects data
dependence between load and store instructions and that detects misspeculation by load instructions”



“mis-speculation” (claims 1, 6, and 9): “when a load instruction that is
dependent for its data on a store instruction appearing earlier in the
program order is in fact executed before the store instruction wrote its data
to a memory address shared with the load instruction”



“in fact executed” (claims 1 and 9): “when a load instruction has actually
accessed a memory address that has not yet been updated by a store
instruction appearing earlier in the program order”



“predictor” (claim 1): “a circuit that receives a mis-speculation indication
from the data speculation circuit to produce a prediction”

D. State of the Prior Art
i.

Overview

By 1995, out-of-order processing was well-known in the field of computer
architecture design. Also by 1995, techniques for detecting data dependence were wellknown in the art. On this much, the parties are in agreement.
Apple further maintains that by 1995, data speculation was well-known in the art,
as were techniques for detecting and recovering from mis-speculations. WARF disputes
10

this, asserting that the prior art techniques do not resemble the solutions proposed by the
’752 patent inventors.

Apple also contends that by 1995, prediction techniques to

improve the accuracy of speculation in an out-of-order processor were well-known in the
art. WARF also disputes this, and in particular contends that the techniques disclosed in
the prior art did not satisfy the “prediction” claimed in the ’752 patent -- the heart of the
parties’ dispute addressed in the opinion below. Finally, Apple also maintains that by
1995, data speculation involving LOAD and STORE instructions was well-known in the
art. WARF disputes that too, arguing the prior art techniques bore no resemblance to
the solutions proposed in the ’752 patent.9
ii.

The Steely Patent

The “Steely patent” is titled “Memory Reference Tagging” and names Simon C.
Steely, Jr., David J. Sager and David B. Fite, Jr. as inventors. The application was filed
on August 12, 1995, and claims priority to an earlier application filed on November 12,
1992. The Steely patent issued as U.S. Patent No. 5,619,662 on April 8, 1997, and was
assigned to DEC. As such, it is prior art to the ’752 patent. Apple contends that the
Steely patent anticipates claims 1-3, 5, 6, and 9 of the ’752 patent.

9

Apple proposes finding of facts about other prior art references, including: a technique
developed by Digital Equipment Corporation (“DEC”); U.S. Patent No. 5,666,506 (“Hesson”);
and a commercial processor known as the Alpha 21264 or “EV6.” As best as the court can
discern, however, these prior art references are only material to Apple’s motion for summary
judgment on the objective prong of WARF’s willful infringement claim. As discussed below, the
court reserves on that based on any arguments not developed fully at summary judgment, waiting
instead to hear the evidence of infringement and invalidity to be introduced during the first phase
of the trial.

11

Pertinent to this anticipation defense, all claims of the ’752 patent require a
“prediction” associated with a LOAD instruction or with a LOAD/STORE pair. Apple
maintains that the Steely patent describes a processor that executes instructions out of
order and uses a prediction to determine whether to allow speculation for LOAD and
STORE instructions. WARF asserts that Steely fails to disclose any “prediction” capable
of receiving ongoing updates -- or even a “prediction” under Apple’s proposed
construction of that claim term.
E. Person of Ordinary Skill in the Art
The parties dispute what characteristics a person of ordinary skill in the art would
possess, though this dispute is not material to the parties’ respective motions for
summary judgment, or at least the reasons for this court’s disposition of those motions.
Apple maintains that for purposes of the ’752 patent, a person of ordinary skill in the art
would have a Ph.D. in electrical engineering, computer engineering or computer science
with a focus on computer architecture or microprocessor design; or an M.S. or B.S.
degree in electrical engineering, computer engineering or computer science with
significant work experience relating to computer architecture or microprocessor design.
WARF maintains that a person of ordinary skill in the art would have at least a
bachelor’s degree in electrical engineering or computer science, and at least three to five
years of experience in computer design and computer architecture. Alternately, WARF
asserts a person of ordinary skill in the art would have a master’s degree in electrical
engineering or computer science, and at least two to three years of experience in

12

computer design and computer architecture. The experience could be derived from either
industry or academia.
F. IPR Decision
Finally, Apple filed a petition with the United States Patent and Trademark
Office’s Patent Trial and Appeal Board (“PTAB”) seeking inter partes review (“IPR”) of all
claims of the ’752 patent. In the petition, Apple argued that claims 1-9 are invalid as
obvious in view of Hesson and Steely, relying on a declaration of its expert Dr. Colwell.
On April 15, 2015, after briefing by Apple and WARF, the PTAB denied Apple’s petition
“as to all challenged claims,” finding that Apple “has not shown . . . that there is a
reasonable likelihood that it will prevail” on its obviousness theory for any claim of the
’752 patent.

(4/17/2015 Declaration of Christopher Abernathy (“4/17/15 Abernathy

Decl.”), Ex. A (dkt. #151-1) p.3.)
In particular, the PTAB construed “prediction” as “a variable that indicates the
likelihood that the data speculative execution of a load instruction will result in a misspeculation.” (Id. at p.10.) In so finding, the PTAB reasoned:
We agree that in the ’752 patent, the mis-speculation
prediction at any point in time is a function of the misspeculation history of load-store instruction pairs. Thus, the
prediction is a variable. The fact that the prediction has a
particular value at each point in time is merely an indication
of its functional relationship and does not change the nature
of the prediction from a variable to a constant value.
(Id.)10

10

In Wisconsin Alumni Research Foundation v. Intel Corp., No. 08-cv-78-bbc (W.D. Wis. filed Feb. 5,
2008), Judge Crabb similarly construed “prediction” in the same patent to mean “a variable that

13

OPINION
I. Claim Construction
“It is a ‘bedrock principle’ of patent law that ‘the claims of a patent define the
invention to which the patentee is entitled the right to exclude.’” Phillips v. AWH Corp.,
415 F.3d 1303, 1312 (Fed. Cir. 2005) (en banc) (quoting Innova/Pure Water, Inc. v. Safari
Water Filtration Sys., Inc., 381 F.3d 1111, 1115 (Fed. Cir. 2004)). The court exclusively
determines claim construction as a matter of law. Markman v. Westview Instruments, Inc.,
517 U.S. 370, 372 (1996). The words of the claims are always the “appropriate starting
point” for proper construction, Comark Commc’ns, Inc. v. Harris Corp., 156 F.3d 1182,
1186 (Fed. Cir. 1998), with the court asking “how a person of ordinary skill in the art
understands a claim term” as an “objective baseline from which to begin claim
interpretation,” Phillips, 415 F.3d at 1313.
“Importantly, the person of ordinary skill in the art is deemed to read the claim
term not only in the context of the particular claim in which the disputed term appears,
but in the context of the entire patent, including the specification.” Id. In fact, “[t]he
best source for understanding a technical term is the specification from which it arose,
informed, as needed, by the prosecution history.” Multiform Desiccants, Inc. v. Medzam,
Ltd., 133 F.3d 1473, 1478 (Fed. Cir. 1998). As the Federal Circuit has recognized,

indicates the likelihood that the data speculative execution of a load instruction will result in misspeculation,” and later clarified that “a ‘prediction’ must be capable of receiving ongoing
updates.” Wis. Alumni Research Found. v. Intel Corp., No. 08-cv-78-bbc, 2008 WL 4279975, at *7
(W.D. Wis. Sept. 18, 2008); Wis. Alumni Research Found. v. Intel Corp., 656 F. Supp. 2d 898, 922
(W.D. Wis. 2009). While the court agrees with Apple that this decision has no binding effect on
this court (Def.’s Opening Br. (dkt. #118) 23), any more than the PTAB’s decision does, the
court will obviously consider Judge Crabb’s and the PTAB’s reasoning and analysis in the opinion
below.

14

however, “there is sometimes a fine line between reading a claim in light of the
specification, and reading a limitation into the claim from the specification.”

Liebel-

Flarsheim Co. v. Medrad, Inc., 358 F.3d 898, 904 (Fed. Cir. 2004) (quoting Comark
Commc’ns, 156 F.3d at 1186-87). “[A]n inherent tension exists as to whether a statement
is a clear lexicographic definition or a description of a preferred embodiment.

The

problem is to interpret claims ‘in view of the specification’ without unnecessarily
importing limitations from the specification into the claims.” E-Pass Techs., Inc. v. 3Com
Corp., 343 F.3d 1364, 1369 (Fed. Cir. 2003).
In addition to intrinsic evidence like the specification and prosecution history, the
Federal Circuit has “authorized district courts to rely on extrinsic evidence, which
‘consists of all evidence external to the patent and prosecution history, including expert
and inventor testimony, dictionaries, and learned treatises.’” Phillips, 415 F.3d at 1317
(quoting Markman v. Westview Instruments, Inc., 52 F.3d 967, 980 (Fed. Cir. 1995)).
“However, while extrinsic evidence ‘can shed useful light on the relevant art,’ [the Federal
Circuit has] explained that it is ‘less significant than the intrinsic record in determining
‘the legally operative meaning of claim language.’” Id. (quoting C.R. Bard, Inc. v. U.S.
Surgical Corp., 388 F.3d 858, 862 (Fed. Cir. 2004)). Accordingly, the court can consider
extrinsic evidence in construing patent claims, but it must do so in the context of the
intrinsic evidence and while keeping in mind the flaws inherent in each type of extrinsic
evidence. Id. at 1318.

15

As previously mentioned, the parties dispute the proper construction of only one
claim term, “prediction,” which appears in claims 1, 2, 3, 5 and 9 of the ’752 patent. 11
The parties propose the following constructions for that term:
“Prediction”
Plaintiff WARF’s Proposed Construction

Defendant Apple’s Proposed Construction

“A variable that indicates the likelihood “A value that indicates the likelihood that
that the data speculative execution of a the data speculative execution of a load
load instruction will result in a mis- instruction will result in a mis-speculation”
speculation”
A “prediction” must be capable of receiving A “prediction” need not be capable of
ongoing updates.
receiving ongoing updates.
The obvious and sole substantive difference between the parties’ competing constructions
is whether the prediction must be capable of change (while implicit in the “value”“variable” dispute, the second sentence of each definition makes that disagreement
explicit). According to WARF, a “prediction” must be able to receive updates -- in other
words, it must be dynamic. Apple, on the other hand, argues that a “prediction” may be
dynamic, but it may also be static -- that is, incapable of receiving ongoing updates and
changing to reflect those updates.
The language of claim 1 reads as follows:
In a processor capable of executing program instructions in an
execution order differing from their program order, the
processor further having a data speculation circuit for
detecting data dependence between instructions and
detecting a mis-speculation where a data consuming
instruction dependent for its data on a data producing
11

The parties also mention a possible dispute about the meaning of “table,” but neither
party sought construction at summary judgment.
16

instruction of earlier program order, is in fact executed before
the data producing instruction, a data speculation decision
circuit comprising:
a) a predictor receiving a mis-speculation indication from the
data speculation circuit to produce a prediction associated
with the particular data consuming instruction and based
on the mis-speculation indication; and
b) a prediction threshold detector preventing data
speculation for instructions having a prediction within a
pre-determined range.
(’752 patent, 14:36-52 (emphasis added).)
The court can dispense with one of Apple’s arguments at the outset. Apple points
out that none of the claims at issue contain an express limitation requiring the prediction
to be updated on an ongoing basis, suggesting that this means predictions need not be
capable of update. (Def.’s Br. Support Summ. J. (dkt. #118) 19-20.) As appealing as
that simple construction might be, since it would alleviate the need for further analysis,
the lack of an express limitation actually requires further inquiry: if the claims contained
an additional limitation requiring a dynamic prediction, construing the term “prediction”
to be intrinsically dynamic would render that limitation superfluous. See Digital-Vending
Servs. Int’l, LLC v. Univ. of Phoenix, Inc., 672 F.3d 1270, 1275 (Fed. Cir. 2012) (discussing
the “well-established rule that claims are interpreted with an eye toward giving effect to
all terms in the claim”) (internal quotation marks and citation omitted); cf. LSI Indus.,
Inc. v. ImagePoint, Inc., 279 F. App’x 964, 972 (Fed. Cir. 2008) (“Some claims specifically
recite ‘an illuminated display device,’ while others recite only ‘a display device.’ . . . Thus,
the language of the claims counsels against imposing an illumination limitation on the
display device term because it would make the limitation superfluous where it explicitly
17

appears.”); Phillips, 415 F.3d at 1314 (“To take a simple example, the claim in this case
refers to ‘steel baffles,’ which strongly implies that the term ‘baffles’ does not inherently
mean objects made of steel.”). Thus, although the claims include no limitations explicitly
requiring predictions to be dynamic, the word “prediction” itself still might (or might
not) include that requirement depending on the claim language, specification,
prosecution history and extrinsic evidence.
WARF relies heavily on the fact that the claimed “data speculation decision
circuit” prevents data speculation for instructions having a prediction “within a predetermined range.” According to WARF, the claimed function of determining whether a
prediction falls within a given range makes sense only if, “at any given time after the
‘prediction’ is produced, it might be ‘within a predetermined range’ or it might not be.”
(Pl.’s Br. Opp’n Summ. J. (dkt. #148) 8 (emphasis in original).)

Apple argues in

response that this interpretation narrows the claims in a way not supported by the text or
the state of the prior art. In particular, Apple points out that prior art in the field,
including an article entitled “A Study of Branch Prediction Strategies” by James E. Smith
(Decl. of Bryan S. Conley, Ex. 12 (dkt. #124-12) [hereinafter “Smith” or the “Smith
article”]), used the word “prediction” in the context of speculation strategies tracking
single past events, rather than a dynamic history of such events.12 According to Apple,

12

The Smith article deals with control speculation, rather than data speculation. Control
speculation involves “branch prediction.” In the words of the ’752 patent, it “might involve
executing an instruction that follows a branch instruction without knowing the outcome of the
branch (and thus whether the following instruction should have been executed or was branched
around).” (’752 patent, 2:32-36.)

18

the claims certainly permit a dynamic prediction but are also broad enough to encompass
a prediction incapable of receiving updates.
In Wisconsin Alumni Research Foundation v. Intel Corp., 656 F. Supp. 2d 898 (W.D.
Wis. 2009), this court relied in part on the same language WARF cites, finding that:
Claim 1 describes “producing” a “prediction” from a “misspeculation indication” generated in a data speculation circuit
and determining whether that “prediction” is “within a
predetermined range” to decide whether to prevent data
speculation. Thus, the claim language itself establishes that a
“prediction” is something other than a stored “indication”
and is capable of having a “range” of values . . . On its face,
this language suggests that a prediction must be capable of
changing over time.
Id. at 922.
Revisiting this same claim language here, the court again finds the contemplated
use of a predetermined range of values to assess whether instructions should be permitted
to speculate favors WARF’s narrower interpretation. By way of example, imagine a pair
of instructions that mis-speculates for the first time. The parties agree that a single misspeculation is enough to produce a prediction; thus, in this instance, the predictor of the
invention would receive that mis-speculation indication from the data speculation circuit
and use it to produce a prediction of “1,” representing the single mis-speculation. Under
Apple’s construction, the development of the prediction can end here, because it need
not be capable of further updates. Thus, the prediction would be set permanently at its
initial value of “1.”
Under this approach, the next time the instructions execute, there is no need for
the prediction threshold detector to assess whether the prediction falls within a “pre19

determined range.” In a static situation, there are only two possibilities: either there is a
prediction with a value of “1,” because the instructions have mis-speculated a single time;
or there is no prediction, because the instructions have not yet mis-speculated and,
therefore, the prediction has not yet been created. Thus, under Apple’s construction, the
question for the prediction threshold detector is a binary determination of whether a
prediction exists at all, rather than whether a prediction “falls within a given range.”
Indeed, there would be no need for “a data speculation decision circuit” in claim 1 at all,
feeding ongoing mis-speculation outcomes, since the “data speculation circuit” itself
would provide the single piece of information required for a static prediction.
Effectively, the prediction threshold detector would prevent data speculation.13
Said another way, Apple’s construction would read out the words “within a predetermined range” from subsection (b) of claim 1, or at least render them superfluous in
the context of “predictions” incapable of receiving updates; in those cases, the prediction
threshold detector would prevent data speculation “for instructions having a prediction,”
full stop.
Of course, as Apple argues, a “range” can consist of a single value, which could
technically allow for a “predetermined range” including only the value 1. Superficially,
this provides some support for Apple’s construction, but it still does not explain why it
would ever be necessary to compare an existing prediction to a range of 1 for so-called
13

Theoretically, it is possible that the prediction of “1” would not fall within the predetermined
range and the instruction would be allowed to execute regardless of the previous mis-speculation.
But if that were so, the invention would appear to serve no purpose, since the prediction would
not prevent speculation and could not change, much less improve, the processor’s performance
over time. Likely for this reason, Apple does not advance this argument, so the court does not
consider it further.

20

static “predictions.”

The choice is still binary -- either there is no prediction or the

prediction is set to 1 -- and so the notion of “comparison” remains a poor fit for the kind
of theoretical static “predictions” Apple posits, regardless of whether the predetermined
range is set to encompass multiple values or a single value.
The remainder of the ’752 patent further supports WARF’s construction. The
brief summary of the invention describes a three-tiered approach for determining when
an instruction should execute. The first tier encompasses instructions with no history of
mis-speculation; they may execute “without further inquiry.” (’752 patent, 3:66.) The
second tier implicates instructions that have previously mis-speculated. At that point,
according to the description, the invention employs a predictor “to determine whether
the instruction should be executed or delayed.” (Id. at 4:1-4.) If the prediction were
static, however, the mere fact of its existence would be enough to prevent execution. In
contrast, the predictor as described in the ‘752 patent instead uses “the past history of
mis-speculations” to determine whether the instruction may execute, allowing those that
are “typically not dependent” to execute immediately. (Id. at 4:1-5 (emphasis added).)
This language, too, suggests a prediction capable of update; it makes little sense to speak
of instructions that are “typically not dependent” when a single instance of misspeculation could, under Apple’s construction, foreclose future speculative execution
without the possibility of updates to reflect what typically occurs.

If the predictor

ultimately delays the instruction, the third tier then employs a synchronization table to
determine when the instruction should execute, delaying it “until after the execution of
the particular data producing instruction” on which it depends. (Id. at 4:5-7, 27-28.)
21

Furthermore, this three-tiered approach appears in the brief summary of the
invention, rather than as a description of a single embodiment, making it more persuasive
as a source of support for WARF’s narrower construction. See C.R. Bard, Inc. v. U.S.
Surgical Corp., 388 F.3d 858, 864 (Fed. Cir. 2004) (“Statements that describe the
invention as a whole, rather than statements that describe only preferred embodiments,
are more likely to support a limiting definition of a claim term. . . . Statements that
describe the invention as a whole are more likely to be found in certain sections of the
specification, such as the Summary of the Invention.”).
While less persuasive given its location in the patent, the detailed description of
the invention provides further context suggesting that a “prediction” must be dynamic.
As this court recognized in describing the preferred embodiment of the invention in Intel,
“the specification explains in unequivocal terms that ‘[t]he prediction provided by the
predictor circuit 33 . . . is updated based on historical mis-speculations detected by the
data speculation circuit 30.

For this reason, the data speculation circuit 30 must

communicate with the predictor circuit 33 on an ongoing basis.’” Intel, 656 F. Supp. 2d at
922 (quoting ’752 patent, 8:7-11) (emphasis added).

WARF also points to other

examples supporting its position in the description of the preferred embodiment,
including the description of the way the prediction normally “is incremented and
decremented” such that “the higher the prediction 109, the more likelihood of misspeculation[.]”

(’752 patent, 11:29-35 (emphasis added).)

Indeed, throughout the

description of the preferred embodiment, the specification consistently refers to the
prediction as dynamic. (See, e.g., id. at 12:14-17 (“[T]he prediction that there was a need
22

to synchronize was wrong and so at process block 120 the prediction 109 is decremented
toward the do not synchronize state.”) (emphasis added); 12:52-54 (“[T]he prediction
109 is updated toward the synchronize condition indicating that the prediction that there
was a need to synchronize was correct[.]”) (emphasis added); 12:67-13:3 (“If [a misspeculation occurs and the pair is already in the prediction table] then at process block
302, the prediction 109 is updated toward synchronize so that this mis-speculation may
be avoided in the future.”) (emphasis added).)
Acknowledging, as it must, that the preferred embodiment describes a dynamic
prediction that receives ongoing updates (Def.’s Br. Support Summ. J. (dkt. #118) 21;
Def.’s Br. Opp’n Summ. J. (dkt. #140) 16 ), Apple relies on the general principle that “it
is improper to read limitations from a preferred embodiment described in the
specification -- even if it is the only embodiment -- into the claims absent a clear
indication that the patentee intended the claims to be so limited.” Liebel-Flarsheim Co.,
358 F.3d at 913. But this is not a case in which the “claim language is sufficiently broad
that it can be read to encompass features not described in the written description, either
by general characterization or by example in any of the illustrative embodiments.” Id. at
905.

Rather, as described above, the claims themselves suggest that the contemplated

“prediction” must be capable of change; the preferred embodiment merely provides
further support for that conclusion.
Use of the preferred embodiment as context, rather than as a source of limitations
that do not otherwise appear in the claims, is permissible. Compare, e.g., Teleflex Inc. v.
Ficosa N. Am. Corp., 299 F.3d 1313, 1327-28 (Fed. Cir. 2002) (district court erred in
23

holding that “clip” was limited to a “single pair of legs,” even where that was the only
embodiment described, where claim language did not support that limitation,
specification and prosecution history included no statements of restriction and the
ordinary meaning of “clip” was not so restricted), with Toro Co. v. White Consolidated
Indus., Inc., 199 F.3d 1295, 1301-02 (Fed. Cir. 1999) (holding that construction of
“including” required attachment between structures where that was the only embodiment
disclosed and where nothing in the remainder of the specification supported an
unattached embodiment; “[T]he specification describes the advantages of the unitary
structure as important to the invention. . . . No other, broader concept was described as
embodying the applicant’s invention, or shown in any of the drawings, or presented for
examination.”).
Still, Apple argues that the ’752 patent does expressly contemplate alternative
embodiments of the invention, pointing out that the detailed description of the invention
states:
It will be understood that the prediction 109 may be
obtained by methods other than simply incrementing it in
value for each speculation as described herein. For example,
various weighting schemes can be provided to cause the
predictor circuit 33, for example, to be less sensitive to the
earliest mis-speculations. More complex pattern matching
techniques may also be used, for example, to catch situations
where mis-speculations occur in groups or regular patterns.
(’752 patent, 14:6-14.) Apple contends that a person of ordinary skill in the art would
read this discussion to allow for alternative embodiments in which a prediction is not
updated on an ongoing basis and urges the court not to “improperly exclude a disclosed
embodiment” by adopting WARF’s construction. See Broadcom Corp. v. Emulex Corp., 732
24

F.3d 1325, 1333 (Fed. Cir. 2013). The problem with this argument is that the patent
simply does not disclose the embodiment Apple advocates.
As a beginning point, neither of the alternative embodiments disclosed in the
specification contemplate a static “prediction.”

To the contrary, both proposed

alternatives implicitly contemplate arrangements involving dynamic predictions. Both
schemes that assign a different weight to later mis-speculations and techniques that
identify mis-speculations occurring in groups or regular patterns assume a developing
history of mis-speculations that the predictor circuit can use to obtain its prediction.
There is no need to weight different instances of mis-speculation if the prediction is static
and will never be updated to reflect those weights.14 There is likewise no need to develop
complex matching techniques to identify patterns in mis-speculation if the prediction can
never take that information into account in determining how likely an instruction is to
mis-speculate. Adopting WARF’s construction, therefore, does not exclude a “disclosed
embodiment” from the scope of the claims.
Nor is there support for Apple’s proposed construction in the intrinsic evidence,
Apple’s arguments to the contrary notwithstanding. For instance, Apple argues that the
specification makes clear that a single mis-speculation is enough to produce a prediction.
(Def.’s Br. Support Summ. J. (dkt. #118) 18.) This is true enough, but it has no bearing
14

Apple uses the mention of “weighting schemes” to propose its own take on an alternative
embodiment as well -- a “weighting scheme that always prevents speculation by a load instruction
for which mis-speculation recovery would be especially costly.” (Def.’s Br. Support Summ. J.
(dkt. #118) 23.) Such a weighting scheme would not, however, be a means of producing a
“prediction.” Both parties agree that a prediction indicates the likelihood that a pair of instructions
will mis-speculate; Apple’s embodiment has nothing to do with the likelihood of mis-speculation,
but rather assesses whether the costs associated with a single mis-speculation are prohibitive,
regardless of how likely or unlikely that mis-speculation is.

25

on whether a prediction must be capable of update after its creation -- nor does the fact
that the claim language “does not specify any minimum number of times that the
instruction must mis-speculate before the ‘prediction’ is above the threshold required to
prevent speculation.” (Id.) Apple also argues that a “prediction” must be construed as a
“value” (i.e., static) rather than a “variable” (i.e., dynamic) because the specification
“explicitly describes the prediction as a ‘value.’” (Id. at 15-16.) But the examples it cites
speak of the prediction being set to a “default value” or being “incremented in value.” All
this confirms is that the prediction is some number that has a value; it does not suggest
the value of that prediction cannot change.

To the contrary, the portions of the

specification Apple cites refer to “incrementing” the value of the prediction, suggesting
that it can and does change. Thus, the court again adopts Judge Crabb’s conclusion in
Intel that “[n]either the claim language nor the specification supports defendant’s
proposed construction that a ‘prediction’ may include values that are fixed once to
indicate a single incident of mis-speculation.” 656 F. Supp. 2d at 922.
Finally, Apple contends that its construction finds support in extrinsic evidence,
citing to the Smith article discussed above, as well as the reports of its two experts, Dr.
August and Dr. Colwell. Both reports, however, primarily rehash Apple’s legal arguments
by purporting to analyze the language in the specification and claims. (See August Report
(dkt. #103) ¶¶ 136-47; Colwell Report (dkt. #104) ¶¶ 131-41.)

The court rejects

Apple’s positions on those issues, and so, too, expert reports that echo those same
arguments.

26

Apple is, therefore, left with the Smith article and another paper, “Memory
Dependence Prediction [U]sing Store Sets,” by George Z. Chrysos and Joel S. Emer (the
“Chrysos paper”) (Conley Decl., Ex. 1 (dkt. #143-1)), both of which Apple contends
describe techniques that produce static predictions. Even if Apple’s characterization were
accurate, these two extrinsic references are wholly underwhelming compared to the
language of the patent itself and contrary intrinsic evidence.

Moreover, extrinsic

evidence “can be used only to help the court come to the proper understanding of the
claims”; it cannot be used to vary or contradict the claim language or specification.
Vitronics Corp. v. Conceptronic, Inc., 90 F.3d 1576, 1584 (Fed. Cir. 1996).
For all these reasons, the court finds WARF’s proposed construction of the term
“prediction” compelling and will construe that term as requiring a prediction that is
capable of receiving updates.
II. Invalidity
A. Anticipation by Steely
On summary judgment, both parties devote most of their invalidity briefing to the
question of whether the ’752 patent is invalid as anticipated by the Steely patent, U.S.
Patent No. 5,619,662. Evaluating a claim of anticipation involves a two-step inquiry.
The first step requires proper construction of the meaning and scope of the claims. Power
Mosfet Techs., L.L.C. v. Siemens AG, 378 F.3d 1396, 1406 (Fed. Cir. 2004). “The second
step in the analysis requires a comparison of the properly construed term to the prior
art[.]”

Id.

To demonstrate anticipation, “the proponent must show ‘that the four

corners of a single, prior art document describe every element of the claimed invention.’”
27

Net MoneyIN, Inc. v. VeriSign, Inc., 545 F.3d 1359, 1369 (Fed. Cir. 2008) (quoting Xerox
Corp. v. 3Com Corp., 458 F.3d 1310, 1322 (Fed. Cir. 2006)). Although anticipation is
ultimately a question of fact, “it may be decided on summary judgment if the record
reveals no genuine dispute of material fact.” Leggett & Platt, Inc. v. VUTEk, Inc., 537 F.3d
1349, 1352 (Fed. Cir. 2008) (quoting Golden Bridge Tech., Inc. v. Nokia, Inc., 527 F.3d
1318, 1321 (Fed. Cir. 2008)).
Both parties have moved for summary judgment of anticipation in light of Steely,
and each relies on its own, preferred construction of the disputed term “prediction.”
(Def.’s Br. Support Summ. J. (dkt. #118) 25-37; Pl.’s Br. Support Summ. J. (dkt. #120)
38-64.)

Having just rejected Apple’s construction of that term, Apple’s motion for

summary judgment will be denied.

Even if the court adopts WARF’s construction,

however, Apple maintains numerous disputed issues of fact preclude entry of summary
judgment against it on grounds of anticipation. (Def.’s Br. Opp’n Summ. J. (dkt. #140)
27.) It is to that question the court now turns.
i.

Background of the Steely Patent

The Steely patent is entitled “Memory Reference Tagging” and describes a
processor that “includes a memory reference tagging store associated with the instruction
scheduler so that the scheduler can reorder memory reference instructions without
knowing the actual memory location addressed by the memory reference instruction.”
(U.S. Patent No. 5,619,662 (dkt. #131-4) Abstract.)

Most relevant to the issue of

anticipation, Steely discloses four different techniques in which a “write buffer” assigns
“memory reference tags” involving a mis-speculation to load and store instructions. Each
28

of those techniques appears in the section of the patent entitled “Memory Reference
Tagging.” (See id. at 47:35-49:8.)
In the first technique, a mis-speculation generates a memory reference tag from a
portion of the address in memory that resulted in the LOAD-STORE collision. (Id. at
48:2-4.) Once that portion is placed in the memory tag store, every time an instruction
is retrieved from memory to be executed, the memory reference tag circuit “will provide
the tag bits to be used by the instruction scheduler.”

(Id. at 48:30-33.)

If the

instructions appear with identical tag bits (indicating a previous mis-speculation), the
instruction scheduler will not reorder them. (Id. at 48:33-36.)
In the second technique, the pair of instructions after a mis-speculation is tagged
not with a portion of the memory address, but instead with “a problem number which
could be a number provided from a counter.” (Id. at 48:55-57.) As a result, “[t]wo
memory reference instructions with the same address and number will not reorder.” (Id.
at 48:57-59.)

“However, if the two memory reference instructions have a different

number, the instructions will reorder.” (Id. at 48:59-61.) The counter does not appear
to increment with respect to that for the same pair of instructions once it has assigned
the “problem number”; rather, it increments when a mis-speculation occurs with respect
to a different pair of instructions. For instance, if a pair of instructions mis-speculates and
is assigned the problem number 0, the next pair to mis-speculate might be assigned the
problem number 1.
The third technique is to assign an instruction a “bit to indicate that an
instruction should not be reordered.” (Id. at 48:62-63.) Thus, using this technique, “for
29

a store that previously caused a problem in the write buffer, the instruction is tagged with
a bit indicating that the ISCHED 38 [instruction scheduler] cannot reorder memory
reference instructions around the Instruction tagged with the bit.” (Id. at 48:63-67.)
The final technique is to “turn off reordering” entirely in response to a misapplication under certain circumstances.

(Id. at 49:4-5.)

For example, the patent

suggests turning off reordering when entering a subroutine, based on the general
observation that “during a subroutine call, there are some initial stores and some exiting
loads” and “[i]t would not be desirable to reorder the exiting loads before the initial
stores.” (Id. at 49:4-9.)
ii.

Analysis

Construing “prediction” as a dynamic (updating) “variable that indicates the
likelihood that the data speculative execution of a load instruction will result in a misspeculation,” the remaining question for deciding the anticipation issue before the court
is whether Steely discloses a prediction that can change over time. In Intel, this court
found that it did not, holding that the requirement of a dynamic prediction was “fatal to
defendant’s contention that the four techniques disclosed in the ’662 patent anticipate
the ’752 patent.” 656 F. Supp. 2d at 922. WARF urges the court to adopt the same
result here, arguing that Steely’s four techniques do not disclose predictions that update
on an ongoing basis.

According to WARF, those techniques simply involve tagging

instructions to reflect a single mis-speculation event, without providing a mechanism to
update those tags. See also Intel, 656 F. Supp. 2d at 922 (“For each [technique], the tag is

30

designed to indicate only that a mis-speculation has occurred, not keep track of misspeculations on an ongoing basis.”).
Unsurprisingly, Apple objects to this characterization.

Apple instead contends

that the outcome of Steely’s tag comparison “can change over time for the same pair of
load and store instructions.” (Def.’s Br. Opp’n Summ. J. (dkt. #140) 29.) According to
Apple’s expert, Dr. Colwell, that can occur if, after the write buffer assigns the tags,
additional mis-speculations involving one of the pair of tagged instructions occur. As an
example, Dr. Colwell presumes a situation in which a load instruction, “Inst 1011,” and
store instruction, “Inst 1007,” have been tagged with the same memory address of
“10010,” such that Steely would prevent speculation.

(Colwell Report (dkt. #104)

¶ 303.) Colwell then posits another situation in which a different load instruction, Inst
1012, also mis-speculates with store instruction Inst 1007:
Another load instruction, for instance Inst 1012, may later be
reordered ahead of store instruction Inst 1007, both Inst
1012 and Inst 1007 accessing the same memory address, this
address ending in a different set of 5 bits, for instance
“00110.” Inst 1007 would then be associated with the tag
“00110,” which would no longer be identical to the tag
“10010” associated with the load instruction Inst 1011.
Because the tags for Inst 1007 and Inst 1011 are no longer
identical, Steely predicts they are not dependent and may
reorder them. Thus, the “prediction” disclosed by Steely is a
“variable” that is “capable of receiving ongoing updates,” as
required by WARF’s proposed construction of the term
“prediction.”
(Id.)
Whether Steely actually discloses this means of “updating” its tags within the four
corners of the patent is certainly open to debate. Apple asks the court to infer as much,
31

based on the fact that: (1) the memory reference tag store is large enough to store just
one tag per instruction; and (2) the patent describes how tags for each mis-speculation
“will be stored” regardless of other tags that may already exist for those instructions.
According to Apple, these two facts demonstrate that Steely overwrites previously stored
tags, thereby “updating” the result of any comparison that Steely performs between the
two.
The flaws in this argument are multiple. Essentially, Apple and its experts assume
what amounts to a defect in Steely, which prevents it from assigning more than one tag
preventing mis-speculation to a single instruction (in Dr. Colwell’s example, Inst. 1007),
even though the example posits tags with different memory addresses (Inst. 10010 and
Inst. 00110), depending on the store instruction with which the load instruction 1007 is
paired (here, Inst. 1011 and Inst. 1012). Not only is this assumption contradicted by
the language in Steely, see U.S. Patent no. 5,619,662, at 47:37-43 (“The memory
reference tag store . . . provides at least one bit associated with said instruction . . .”)
(emphasis added), but it would undermine the whole purpose of Steely, which is to
prevent future mis-speculations, since it would result in a never-ending loop for load
instructions causing multiple mis-speculations each time the 10010 and 00110 tags
overwrite one another.15

15

The inventor of Steely does appear to have admitted in his deposition that in his view, this is
how his invention would function, though he was asked the question out of context and without
being asked about the obvious defect this would appear to create in his patent. (See Def.’s Br.
Opp’n Summ. J. (dkt. #140) 31.) As WARF points out, this is why after-the-fact testimony of
the inventor is of limited relevance when unsupported by the patent itself. See Howmedica
Osteonics Corp. v. Wright Med. Tech., Inc., 540 F.3d 1337, 1346 (Fed. Cir. 2008) (“The testimony
of an inventor ‘cannot be relied on to change the meaning of the claims.’”) (quoting Markman, 52

32

More importantly for summary judgment purposes, even assuming one might infer
that such overwriting occurs and constitutes “updating,” the above-described “tag
replacement system” would hardly constitute a “prediction” as this court has construed
the term. Properly construed, a “prediction” must communicate the likelihood of misspeculation and must be capable of update. Using the example offered by Dr. Colwell for
the sake of simplicity, Inst 1011 and Inst 1007 proved to be dependent and were,
accordingly, tagged with the same memory address. Thereafter, another load instruction,
Inst 1012, also proves to be dependent on store instruction Inst 1007. Accordingly,
Steely overwrites the first tag on Inst 1007, tagging it to match Inst 1012, but that is not
so much an “update” of the comparison between Inst 1007 and Inst 1011 as it is the
wholesale elimination of that comparison. By Apple’s and Dr. Colwell’s own description,
no record of the previous mis-speculation remains; the next time the tags are compared
under Steely, they fail to reflect that any mis-speculation has occurred in the past and,
therefore, fail to communicate the likelihood that the data speculative execution of the
load instruction 1011 and store instruction 1007 will result in a mis-speculation. In
contrast, the invention of the ’752 patent incrementally increases the prediction for each
mis-speculation associated with an instruction pair, while it decrements the prediction
associated with a pair of instructions when they do not mis-speculate, thereby updating its
assessment of the likelihood of mis-speculations in the future.
Steely’s tag replacement system, even as explained by Dr. Colwell, discards the
prediction associated with a pair of instructions when a different pair of instructions misF.3d at 983). What matters for purposes of anticipation is what the patent actually discloses, not
what the inventor says it would do in a situation the patent does not clearly address.

33

speculates. While this data elimination admittedly yields a change in the result of the tag
comparison, that change has nothing to do with updating the likelihood that the first pair
of instructions will mis-speculate again in the future. Accordingly, no reasonable jury
could find that the Steely patent discloses each and every limitation of the ’752 patent as
properly construed. The court will, therefore, grant summary judgment to WARF on
defendant’s Steely anticipation defense and counterclaim.

B. Indefiniteness
Finally, Apple contends that claims 5 and 6 of the ’752 patent are invalid as
indefinite.

“[A] patent is invalid for indefiniteness if its claims, read in light of the

specification delineating the patent, and the prosecution history, fail to inform, with
reasonable certainty, those skilled in the art about the scope of the invention.” Nautilus,
Inc. v. Biosig Instruments, Inc., 134 S. Ct. 2120, 2124 (2014).

A party raising an

indefiniteness challenge, like other invalidity challenges, bears the burden of proving that
invalidity by clear and convincing evidence. Microsoft Corp. v. i4i Ltd. P’ship, 131 S. Ct.
2238, 2242 (2011); see also 35 U.S.C. § 282.
Here, Apple contends that claims 5 and 6 of the ’752 patent should be held
indefinite under Nautilus solely because certain terms in those claims lack an antecedent
basis. Claim 5 is a dependent claim and reads:
The data speculation decision circuit of claim 2 wherein the
instruction synchronization circuit includes a synchronization
table associating the certain data consuming instructions and the
certain data producing instructions each with a flag value
indicating whether the respective certain data producing
instruction has been executed and wherein the instruction
34

synchronization circuit delays the particular data consuming
instruction only:
i)

when the prediction associated with the data
consuming instruction is within a predetermined
range; and

ii)

when the particular data consuming instruction is in
the prediction table; and

iii)

when the flag indicates the particular data producing
instruction has not been executed.

(’752 patent, 15:7-20 (emphasis added).)

Similarly, Claim 6 likewise depends from

claim 2 and reads:
The data speculation decision circuit of claim 2 wherein the
instruction synchronization circuit creates an entry in the
synchronization table including the particular data consuming
instructions and data producing instructions and the flag value
only after a mis-speculation indicating is received for the
particular data consuming instruction and the particular data
producing instruction.
(’752 patent, 15:21-27 (emphasis added).) Apple focuses on the italicized portions of
each of the above claims in making its § 112 argument.
According to Apple, the use of the definite article “the” in each of the above
italicized instances suggests that the terms that article introduces must refer to specific
claim elements already previously discussed. (Def.’s Br. Support Summ. J. (dkt. #118)
39.) See also, e.g., Warner-Lambert Co. v. Apotex Corp., 316 F.3d 1348, 1356 (Fed. Cir.
2003) (“[I]t is a rule of law well established that the definite article ‘the’ particularizes
the subject which it precedes. It is a word of limitation as opposed to the indefinite or
generalizing force of ‘a’ or ‘an.’”) (quoting Am. Bus Ass’n v. Slater, 231 F.3d 1, 4-5 (D.C.
Cir. 2000)). As Apple points out, the italicized terms above do not appear elsewhere in
35

claims 5 and 6 themselves, or in claims 1 and 2, on which both claims 5 and 6 ultimately
depend. In Apple’s view, this makes it impossible for a person of skill in the art to
determine the scope of claims 5 and 6, rendering them indefinite.
In Halliburton Energy Services, Inc. v. M-I LLC, 514 F.3d 1244 (Fed. Cir. 2008), the
Federal Circuit held that “a claim could be indefinite if a term does not have proper
antecedent basis where such basis is not otherwise present by implication or the meaning
is not reasonably ascertainable.” Id. at 1249; see also Energizer Holdings, Inc. v. Int’l Trade
Comm’n, 435 F.3d 1366, 1370 (Fed. Cir. 2006) (citing Slimfold Mfg. Co. v. Kinkead Indus.,
Inc., 810 F.2d 1113, 1116 (Fed. Cir. 1987)). The specification can, however, provide
sufficient context for a person skilled in the field of the art to understand the claim to
render it definite. See, e.g., In re Skvorecz, 580 F.3d 1262, 1268 (Fed. Cir. 2009) (“We
agree with Mr. Skvorecz that the clause ‘welded to said wire legs at the separation’ does
not require further antecedent basis in claim 1, for a person skilled in the field of the
invention would understand the claim when viewed in the context of the specification.”).
Here, the terms in question are “reasonably ascertainable” in light of the patent’s
specification.
Taking first the terms “the certain data consuming instructions” and “the certain
data producing instructions” in claim 5, the patent’s specification summarizes the
invention and notes that the invention’s instruction synchronization circuit:
may also include a synchronization table associating certain
data consuming instructions and certain data producing instructions,
each with a flag indicating whether the respective data
producing instruction has been executed. The instruction
synchronization circuit delays the subsequent instances of the
certain data consuming instruction only when the prediction
36

associated with the data consuming instruction is within a
predetermined range and when the particular data consuming
instruction is in the prediction table and when the flag
indicates that particular data producing instruction has not
been executed.
(’752 patent, 4:54-65 (emphasis added).)

As WARF points out, this portion of the

specification tracks the language of claim 5 almost exactly. There is no reason why a
person of ordinary skill in the art would not read “the certain data consuming
instructions” and “the certain data producing instructions” to be those included in the
synchronization table in light of the specification. At the very least, the brief summary of
the invention allows one skilled in the art to proceed with “reasonable certainty,” as
Nautilus requires.
The term “the prediction table” in subsection (ii) of claim 5 would likewise inform
a person of ordinary skill in the art that the “prediction table” is contained in the
instruction synchronization circuit. As the brief summary of the invention states, “[t]he
instruction synchronization circuit may include a prediction table listing certain data
consuming instructions and certain data producing instructions each associated with a
prediction.” (’752 patent, 4:39-42 (emphasis added).) The instruction synchronization
circuit then employs the entries in that prediction table in determining whether to delay
subsequent instances of the data consuming instruction -- the instruction must be in the
prediction table for delay to take place. (Id. at 4:48-53.)
As for claim 6, the “synchronization table” is the one that “may” be included in
the instruction synchronization circuit (which is explicitly claimed in independent claim
2) and “associate[s] certain data consuming instructions and certain data producing
37

instructions, each with a flag indicating whether the respective data producing instruction
has been delayed.” (’752 patent, 4:54-58.) The “flag value” likewise takes its meaning
from this portion of the specification, which indicates that each pair of instructions in the
synchronization table has “a flag indicating whether the respective data producing
instruction has been executed.” (See id.) The invention then uses the flag to determine
when to delay execution of subsequent instances of the data consuming instruction. (Id.
at 4:58-65.) A person of ordinary skill in the art would understand the scope of “the flag
value” in claim 6 in light of this relatively clear context. (Id.)
Importantly, because Apple does not dispute that the specification offers context
for the claim terms it identifies, that argument is waived. See Fresenius USA, Inc. v. Baxter
Int’l, Inc., 582 F.3d 1288, 1296 (Fed. Cir. 2009) (“If a party fails to raise an argument
before the trial court, or presents only a skeletal or undeveloped argument to the trial
court, we may deem that argument waived on appeal.”); Jordan v. Binns, 712 F.3d 1123,
1134 (7th Cir. 2013) (undeveloped arguments considered waived); Ultratec, Inc. v.
Sorenson Commc’ns, Inc., No. 13-cv-346-bbc, 2014 WL 3565409, at *1 (W.D. Wis. July
17, 2014). Regardless, Apple takes an entirely different tack, one which requires a bit of
explanation. According to Apple, in light of the antecedent basis problems in claims 5
and 6, a person of ordinary skill in the art might simply look to the specification to
understand the scope of the invention. However, Apple argues, she might also assume
that claims 5 and 6 do not, in fact, depend from claim 2 but instead were intended to
depend from claims 3 and 5, respectively, which would provide the requisite antecedent

38

basis for the identified terms, but would also include additional limitations by virtue of
depending from different claims. (Def.’s Br. Opp’n Summ. J. (dkt. #140) 41.)
The court does not find Apple’s argument persuasive. Apple cites no cases in
which courts found indefiniteness due solely to a lack of antecedent basis, at least where
the specification so clearly delineates the structure of what the patent intended to claim.
Instead, Apple cites Novo Industries, L.P. v. Micro Molds Corp., 350 F.3d 1348 (Fed. Cir.
2003), for the proposition that claims are indefinite where “in light of the mistakes in the
claims there is no clear choice as to how to interpret their scope.” (Def.’s Br. Opp’n
Summ. J. (dkt. #140) 46.) But Novo involved an obvious typographical error amenable
to no fewer than four possible interpretations (at least one of which would have
significant substantive implications for the scope of the claims).16 Novo does not support
this court reading in a typographical error to create ambiguity where the specification
otherwise indisputably provides context to delineate the scope of the invention “with
reasonable certainty.” Nautilus, 134 S. Ct. at 2124.
The other case Apple cites, Automed Technologies, Inc. v. Microfil, LLC, 244 F. App’x
354 (Fed. Cir. 2007), is similarly unhelpful to its indefiniteness argument. In Automed,
the Federal Circuit vacated and remanded a grant of summary judgment of non-

16

In Novo, the claim included a “stop means formed on a rotatable with said support finger.” 350
F.3d at 1352 (emphasis removed). Novo suggested correcting the claim either by deleting the
words “a rotatable with” or by deleting the words “with said.” Id. at 1357. The district court
raised another possibility by changing the word “a” to “and.” Id. And Micro Molds proposed as a
fourth possibility that a word, such as “skirt” or “disk,” might have been erroneously omitted,
which would add an additional substantive limitation to the claims. Id. Because the Federal
Circuit “[could not] know what correction [was] necessarily appropriate or how the claim should
be interpreted,” it concluded that the claim was necessarily indefinite “in its present form.” Id.
No comparable indefiniteness is even arguable in this case.

39

infringement because the district court had based its ruling on a finding that the accused
systems lacked a “controller” -- a limitation that was actually absent from the asserted
claims. Id. at 359. In the midst of that discussion, the Federal Circuit observed:
We also note that claim 27 of the ‘671 patent, which recites
“the controller,” appears to be mistakenly dependent on claim
20, in which this term finds no antecedent basis. . . . Because
claim 21 - and not claim 20 - recites a “controller” limitation,
perhaps claim 27 was intended to depend from claim 21.
Id. Even so, the Federal Circuit said nothing about that potential error rendering claim
27 indefinite. Rather, it “[left] to AutoMed any corrective action it deem[ed] necessary.”
Id. The Federal Circuit’s observation that claim 27 might have been intended to depend
from claim 21, not claim 20, certainly does not compel, or even do much to support, a
finding of indefiniteness in this case.
Accordingly, the court finds that the specification provides ample guidance as to
what elements the claims are referencing when they refer to “the certain data consuming
instructions,” “the certain data producing instructions” and “the prediction table” (claim
5), as well as “the synchronization table” and “the flag value” (claim 6).

Even the

authority upon which defendants rely indicates that a lack of antecedent basis renders a
claim indefinite only if “it would be unclear as to what element the limitation was
making reference.” Manual of Patent Examining Procedure § 2173.05(e) (9th ed. 2014);
see also Halliburton, 514 F.3d at 1249. That is simply not the case here.

40

III. Willful Infringement

WARF has alleged a claim that Apple’s infringement was willful, thereby
permitting (but not requiring) the court to award enhanced damages. 35 U.S.C. § 284
(“[T]he court may increase the damages up to three times the amount found or
assessed.”); Beatrice Foods Co. v. New Eng. Printing & Lithographing Co., 923 F.2d 1576,
1578 (Fed. Cir. 1991) (“It is well-settled that enhancement of damages must be premised
on willful infringement or bad faith.”) (citations omitted).

Apple seeks summary

judgment on this claim on the basis that WARF cannot as a matter of law meet the
threshold for proving willfulness on an objective basis.
To establish willful infringement, WARF “must show by clear and convincing
evidence” (1) that “the infringer acted despite an objectively high likelihood that its
actions constituted infringement of a valid patent,” and (2) that “this objectively-defined
risk . . . was either known or so obvious that it should have been known to the accused
infringer.” In re Seagate Tech., 497 F.3d at 1371. The former “objective determination of
recklessness” is a question for the court, not the jury. Bard Peripheral Vascular, Inc. v.
W.L. Gore & Assocs., Inc., 682 F.3d 1003, 1006-07 (Fed. Cir. 2012).
“[T]he ‘objective’ prong of Seagate tends not to be met where an accused infringer
relies on a reasonable defense to a charge of infringement.”

Id. at 1005-06 (internal

citation and quotation marks omitted); see also Spine Solutions, Inc. v. Medtronic Sofamor
Danek USA, Inc., 620 F.3d 1305, 1319 (Fed. Cir. 2010) (overturning jury’s finding of
willful infringement, finding that defendant raised a “substantial question as to the
obviousness” of the patent in suit); Douglas Dynamics, 747 F. Supp. 2d at 1112 (granting
41

summary judgment on willful infringement claim where there was “reasonable difference
of opinion” and a “close question”).
In cursory fashion, Apple’s opening brief advances a wide range of arguments for
seeking summary judgment on this objective prong. Some of the bases were fully briefed
for review on the merits – namely, Apple’s claim construction of “prediction,” its related
argument on anticipation by Steely and its indefiniteness defense and counterclaim as to
claims 5 and 6. The court will take up Apple’s motion on these bases in the discussion
below.
Other bases, including ones on which Apple bears the burden of proof like
obviousness, were not, however, the subject of the parties’ motions for summary
judgment. While the court appreciates that it is WARF’s burden to demonstrate that
Apple’s defenses to infringement or claims of invalidity are not objectively reasonable,
Apple’s scattershot approach in its motion renders the task near impossible to resolve on
summary judgment.

Perhaps if Apple had identified two or three of its strongest

arguments, this may have been a manageable task. Instead, Apple’s treatment of each
basis is limited to a paragraph or two in its opening brief and reflects ships passing in the
night in reply to WARF’s responses.17

In any event, WARF did come forward with

evidence and law that, despite Apple’s attempt to refute it in reply, could lead to a
finding that Apple’s belief that it either did not infringe the ’752 patent or that the

17

Perhaps most telling, the few defenses that Apple moved on the merits do not offer grounds for
the court to find for Apple on the objective prong of WARF’s willful infringement claim.

42

patent was invalid was not objectively reasonable.18 As such, the court will await a more
robust demonstration of the merits of Apple’s defenses and WARF’s infringement claims
at trial.19
Returning to those bases which were fully briefed for review on the merits, Apple’s
claim construction is arguably “objectively reasonable” if viewed purely in a vacuum.
Apple presented some evidence that “prediction” can describe a static value in the context
of computer circuits and speculation, for example, in the form of the Smith article and
Chrysos paper; they are also correct that the patent does not explicitly define “prediction,”
ostensibly leaving at least some room for debate. The problem is that nothing in the
patent -- not the claim language, not the specification, not the purpose of the invention -supports Apple’s construction. As discussed above, the claim language from the outset
suggests that a prediction must be dynamic in the context of this particular invention.
The specification, including both the brief summary of the invention and the detailed
description of the embodiments, further supports this construction. And Apple’s resort
to extrinsic evidence fails to render its arguments to the contrary any more reasonable,

18

Certainly, Apple seems to make an objectively reasonable argument as to claims 1 and 2 being
obvious, and perhaps as to claims 3 and 9, but the court cannot say on this record whether the
supposed links drawn between Steely, Hesson, Chen and EV6 are obvious or pure sophistry.
Similarly, while Apple raises a number of arguments that appear to objectively establish noninfringement on a literal basis, it has left the court unconvinced as to infringement under the
doctrine of equivalents.
19

To clarify, while the jury is deliberating on liability, the court can take up any additional
evidence and argument relevant to the objective prong of WARF’s willful infringement claim and
likely will render a decision on the objective prong before the parties present any evidence on the
subjective prong during the second phase of the trial (assuming the jury finds infringement and
does not find invalidity).

43

given that extrinsic evidence cannot be used to vary the intrinsic evidence under settled
principles of claim construction.
While superficially appealing, not unlike a siren’s song, Apple’s construction
crashes against the rocks of the patent language itself and intrinsic evidence. Given how
strongly the patent itself supports WARF’s narrower construction, and how little Apple
has to offer in support of its broader one, Apple’s position is not objectively reasonable.
Compare Cohesive Techs., Inc. v. Waters Corp., 543 F.3d 1351, 1374 (Fed. Cir. 2008) (no
willfulness where disputed term “was susceptible to a reasonable construction under
which [the] products did not infringe”), with SSL Services, LLC v. Citrix Systems, Inc., 769
F.3d 1073, 1091 (Fed. Cir. 2014) (affirming district court’s finding of willful
infringement, in part, because defendant’s non-infringement defense based on an
unwarranted limitation of a claim term was not objectively reasonable); cf. Raylon, LLC v.
Complus Data Innovations, Inc., 700 F.3d 1361, 1369 (Fed. Cir. 2012) (finding position on
claim construction frivolous under Rule 11 where proffered construction was “contrary to
all the intrinsic evidence and does not conform to the standard canons of claim
construction”). Finally, while Judge Crabb granted summary judgment to the defendant
in Intel on WARF’s willful infringement claim, she did so on a basis unrelated to claim
construction and one not before this court.

Intel, 656 F. Supp. 2d at 924 (finding

defendant’s licensing defense objectively reasonable). Accordingly, the court will deny
defendant’s motion for summary judgment on plaintiff’s willful infringement claim that
depends on Apple’s claim construction, finding this defense objectively unreasonable.

44

As for Apple’s anticipation challenge to the validity of the ’752 patent based solely
on Steely, the court finds this defense not objectively reasonable as well, though it will
reserve on any obviousness defense involving Steely.

Much of Apple’s anticipation

argument depended upon its claim construction, which was not objectively reasonable as
discussed above. Admittedly, Apple attempted to maintain an anticipation defense even
under WARF’s construction, but its dependence on Steely’s purported, defective “tag
overwriting” scheme is likewise unreasonable, given that this dubious overwriting defect
would simply dispose of previous predictions, rather than “updating” them to reflect an
increased likelihood of future mis-speculation.
The court will also deny Apple’s motion with respect to its indefiniteness defense.
Apple points to no case in which a lack of antecedent basis led to a finding of
indefiniteness despite clear context providing that basis in the specification. Even the case
law Apple cites explain that there is no invalidity for indefiniteness so long as the
antecedent basis is present by implication, and Apple waived any contention that the
specification did not serve to provide such context. As a whole then, this defense was not
objectively reasonable, and Apple cannot use it to escape the possibility of enhanced
damages.

ORDER
IT IS ORDERED that:
1) Plaintiff Wisconsin Alumni Research Foundation’s construction of the
disputed term “prediction” is ADOPTED as described in this opinion.

45

2) Defendant and counter claimant Apple, Inc.’s motion for summary judgment
(dkt. #116) is DENIED as to its counterclaims and defenses of anticipation by
Steely and indefiniteness, and DENIED as to plaintiff’s willful infringement
claim premised on (1) Apple’s claim construction, (2) anticipation by Steely,
and (3) indefiniteness of claims 5 and 6. The court RESERVES on the motion
in all other respects.
3) Plaintiff’s motion for summary judgment (dkt. #117) is GRANTED.
Entered this 5th day of August, 2015.
BY THE COURT:

/s/
__________________________________
WILLIAM M. CONLEY
District Judge

46