NOTE: This disposition is nonprecedential.
United States Court of Appeals
for the Federal Circuit
______________________
RAMBUS INC.,
Appellant,
v.
Teresa Stanek Rea, ACTING DIRECTOR, UNITED
STATES PATENT AND TRADEMARK OFFICE,
Appellee.
______________________
2012-1480
(Reexamination No. 95/001,166)
______________________
Appeal from the United States Patent and Trademark
Office, Board of Patent Appeals and Interferences.
______________________
Decided: June 28, 2013
______________________
JOHN M. WHEALAN, of Chevy Chase, Maryland, argued
for appellant. Of counsel on the brief were JEFFREY A.
LAMKEN and MICHAEL G. PATTILLO, JR., MoloLamken,
LLP, of Washington, DC; and KARA F. STOLL, Finnegan,
Henderson, Farabow, Garrett & Dunner, LLP, of Wash-
ington, DC.
2 RAMBUS INC. v. REA
COKE M. STEWART, Associate Solicitor, Office of the
Solicitor, United States Patent and Trademark Office, of
Alexandria, Virginia, argued for appellee. With her on
the brief were RAYMOND T. CHEN, Solicitor, NATHAN K.
KELLEY, Deputy Solicitor, and WILLIAM LAMARCA, Associ-
ate Solicitor.
______________________
Before RADER, Chief Judge, O’MALLEY, and WALLACH,
Circuit Judges.
Opinion for the court filed by Chief Judge RADER.
Dissenting opinion filed by Circuit Judge WALLACH.
RADER, Chief Judge.
This appeal arises from an inter partes reexamination
proceeding before the United States Patent and Trade-
mark Office (PTO). The Board of Patent Appeals and
Interferences (Board) affirmed the Examiner’s rejection of
all 25 claims of U.S. Patent No. 7,287,109 (the ’109 Pa-
tent) under 35 U.S.C. § 102(a). Because substantial
evidence does not support the Board’s conclusion, this
court reverses.
I.
The technology at issue involves methods of control-
ling a dynamic random access memory device (DRAM).
Storing information in a DRAM is referred to as a “write”
operation, and retrieving information from a DRAM is
referred to as a “read” operation. Information is written
to, or read from, a DRAM using a memory controller. For
example, in a write operation, a memory controller
transmits control information to the DRAM, which in-
cludes a write request and “address” information indicat-
ing where (by row and column) the data will be stored.
The DRAM receives the control information and executes
the write operation, i.e., writes the data to the correspond-
ing location. Data and control information are trans-
RAMBUS INC. v. REA 3
ferred between the memory controller and the DRAM
through a set of lines called a “bus.”
Prior to 1990, DRAMs operated “asynchronously” to
the memory controller—that is, read and write operations
were not conducted with reference to a system clock. A
DRAM would execute read/write operations as quickly as
possible after receiving control information from the
memory controller. Between the time the memory con-
troller transmitted the control information and the time
the DRAM executed the operation, the memory controller
had to wait for the DRAM to complete internal preparato-
ry functions. This interface between the memory control-
ler and the DRAM rendered the bus unusable because it
would freeze in a “wait state” where the memory control-
ler could not perform any other functions until the
read/write operation was executed.
As processor speeds increased, DRAMs could not keep
up, which created a bottleneck that diminished the ad-
vantages of faster processors. The 1990s, however, intro-
duced the synchronous DRAM, which executed read/write
operations with reference to a system clock. After the
memory controller transmitted a read/write request to the
DRAM, the DRAM would wait a pre-determined number
of clock cycles before executing the operation. This clock
function eliminated the “wait state” and enabled the
memory controller to perform other tasks and send addi-
tional requests to other DRAMs while the operation was
pending. This process became known as “interleaving.”
This case features different methods of interleaving.
II.
Rambus is the owner of the ’109 Patent, entitled
“Method of Controlling a Memory Device Having a
Memory Core.” It claims priority to U.S. Patent Applica-
tion No. 08/545,292, filed on October 19, 1995, by Richard
M. Barth, et al. In 2009, the PTO granted NVIDIA Cor-
poration’s request for inter partes reexamination of
4 RAMBUS INC. v. REA
the ’109 Patent. The Examiner rejected all 25 claims as
anticipated by an earlier Rambus patent, U.S. Patent No.
6,584,037 (Farmwald). The Board affirmed the rejection
and Rambus sought rehearing. Meanwhile, Rambus and
NVIDIA settled, and NVIDIA withdrew from the proceed-
ings. The Board denied rehearing, and Rambus appealed
to this court where the sole issue on appeal is whether
substantial evidence supports the Board’s finding that
the ’109 Patent is anticipated by Farmwald.
The ’109 Patent discloses and claims certain methods
of controlling data transfers to and from a DRAM. As the
Board found, the specification describes “at least two
embodiments,” which the Board referred to as the “strobe”
embodiment and the “non-strobe” embodiment. J.A. 3;
see ’109 Patent col. 10 ll. 25–67. In both embodiments, a
memory controller transmits a request packet to the
DRAM. ’109 Patent col. 10 ll. 25–67. The request packet
contains control information indicating whether the
DRAM will perform a read or a write operation. J.A. 3–4.
In the strobe embodiment, the memory controller then
transmits a separate “strobe signal” to the DRAM, which
causes the DRAM to execute the operation immediately
(with a minimal inherent delay). J.A. 4; ’109 Patent col. 8
l. 63–col. 9 l. 7; ’109 Patent col. 9 ll. 41–46. As the Board
noted, the benefit of this invention “stems from latency
minimization, resulting in a relatively free data bus for
other data transfers—i.e., the command control infor-
mation tells the [DRAM] to pre-fetch the desired data . . .
and the [DRAM] then waits for the strobe signal to send
the data.” J.A. 4.
In the non-strobe embodiment, the memory controller
“varies the timing of data transmission without use of the
above-described strobe signal.” J.A. 5. Rather, “the
[request] packet contains a delay value that indicates to
the DRAM when the [operation should be executed]
relative to the time at which the request packet is sent.”
RAMBUS INC. v. REA 5
J.A. 4–5 (quoting ’109 Patent col. 10 ll. 52–60). For ex-
ample, if the request packet contained a write request and
a delay value of eight clock cycles, the DRAM would
execute the write operation after eight clock cycles elapse.
As the Board noted, the benefit here is that the “[memory]
controller is able to dynamically adjust the operative
interleave.” J.A. 6 (emphasis added).
III.
The Farmwald patent, titled “Memory Device which
Samples Data after an Amount of Time Transpires,”
likewise discloses and claims a method of controlling data
transfers to and from a DRAM. As the Board found,
Farmwald teaches using a memory controller to send a
request packet to the DRAM, which contains control
information specifying a read or write operation. J.A. 8.
Farmwald’s request packet also contains a bit that selects
an “access-time register” within the DRAM. J.A. 8;
Farmwald col. 9 l. 54–col. 10 l. 1. The access-time regis-
ters are central to Farmwald’s invention.
Each of Farmwald’s DRAMs contains “access-time
registers 173 which store a set of one or more delay times
at which the device can or should be available to send or
receive data.” Farmwald col. 6 ll. 40–45. For example,
one access-time register might contain a delay value of
two system clock cycles, another four system clock cycles,
and another eight, and so forth. The access-time registers
“can be modified and preferably are set as part of an
initialization sequence that occurs when the system is
powered up or reset.” Farmwald col. 6 ll. 45–47. In other
words, the delay values are stored in the access-time
registers on start-up, prior to the memory controller
transmitting any request packets to the DRAM. J.A.
10993 (“Farmwald relies upon the previously received
[delay] value signal . . . .”).
As the Board found, Farmwald teaches that after the
memory controller transmits a request packet to the
6 RAMBUS INC. v. REA
DRAM, the DRAM will wait to execute the specified
read/write operation depending on which access-time
register is selected by the bit in the request packet. J.A.
8; Farmwald col. 9 ll. 23–25 (“The time after which a data
block is driven onto the bus lines is selected from values
stored in slave access-time registers.”). For example, if
the request packet contains a bit that selects access-time
register no. 4, and access-time register no. 4 was previous-
ly programmed with a delay value of eight clock cycles,
the DRAM will execute the operation contained in the
request packet after eight clock cycles elapse.
IV.
The ’109 Patent and Farmwald are no strangers to
one another. First, the ’109 Patent specification distin-
guishes Farmwald as prior art. For example, the ’109
Patent describes Farmwald’s method of controlling a
DRAM as “inflexible” because the start of an operation is
tied to “a predetermined number of clock cycles” after the
request packet is transmitted, and “the number of clock
cycles . . . may be determined by a value stored in a
register within the DRAM.” ’109 Patent col. 10 ll. 32–37.
Moreover, Farmwald’s parent patent, U.S. Patent No.
5,319,755 (Farmwald ’755), shares the same specification
as Farmwald and was cited by the Examiner during
initial examination of the ’109 Patent application. The
Examiner rejected the ’109 Patent claims as being antici-
pated by Farmwald ’755, but withdrew the objection after
Rambus responded to the Office Action.
Further, the ’109 Patent has been widely litigated in
at least seven causes of action. In at least two of these
actions, Farmwald or Farmwald ’755 were directly in
front of the tribunal. In ITC Investigation No. 337-TA-
661, the Respondents and Staff Attorney advanced simi-
lar arguments as the Board does in the present case, but
the Commission affirmed the ALJ’s conclusion that
Farmwald ’755 does not anticipate the ’109 Patent.
RAMBUS INC. v. REA 7
Certain Semiconductor Chips Having Synchronous Dy-
namic Random Access Memory and Prods. Containing
Same, Inv. No. 337-TA-661, 2011 WL 6016982 (October 1,
2011). Likewise, in ITC Investigation No. 337-TA-753,
the ALJ held in an Initial Determination that Farmwald
does not anticipate the ’109 Patent. Certain Semiconduc-
tor Chips and Prods. Containing Same, Inv. No. 337-TA-
753, 2012 WL 927056 (March 2, 2012).
Notwithstanding the above, the Board found in the
present case that Farmwald anticipates the ’109 Patent.
In other words, a reference that was overcome during
initial examination, distinguished as prior art in the
patent specification, and found not to anticipate by at
least two different tribunals, now—according to the
Board—discloses each and every element of the claims.
The result is somewhat surprising, especially with noth-
ing in the record indicating whether the Board considered
these prior decisions. Upon comparison of the relevant
claims, this court concludes that Farmwald does not
anticipate the ’109 Patent’s claims.
V.
This court reviews the Board’s factual findings for
substantial evidence. In re Gartside, 203 F.3d 1305, 1316
(Fed. Cir. 2000). Anticipation is a question of fact. In re
Baxter Travenol Labs., 952 F.2d 388, 390 (Fed. Cir. 1991).
“A claim is anticipated only if each and every element as
set forth in the claim is found . . . in a single prior art
reference.” Verdegaal Bros., Inc. v. Union Oil Co. of Cal.,
814 F.2d 628, 631 (Fed. Cir. 1987). This court reviews the
Board’s claim construction de novo. Cybor Corp. v. FAS
Techs., Inc., 138 F.3d 1448, 1456 (Fed. Cir. 1998) (en
banc).
The parties agree that claim 1 is representative of all
claims:
8 RAMBUS INC. v. REA
1. A method of controlling a memory device hav-
ing a memory core, wherein the method compris-
es:
Providing control information to the memory de-
vice, wherein the control information includes a
first code which specifies that a write operation be
initiated in the memory device;
Providing a signal to the memory device, wherein
the signal indicates when the memory device is to
begin sampling write data, wherein the write data
is stored in the memory core during the write op-
eration;
Providing a first bit of the write data to the
memory device during an even phase of a clock
signal; and
Providing a second bit of the write data to the
memory device during an odd phase of the clock
signal.
’109 Patent col. 41 l. 61–col. 43 l. 2. The central issue is
whether Farmwald discloses the step of “providing a
signal to the memory device, wherein the signal indicates
when the memory device is to begin sampling write data.”
The Board construed the term “signal” broadly to
mean “the physical representation of data,” and concluded
that Farmwald discloses such a signal. The Board’s
construction of the term “signal” is unreasonably over-
broad. Indeed, construing the term “signal” to mean any
representation of data would essentially convert every
element of claim 1 into a signal because the control infor-
mation, including the write command and the data being
written, are all physical representations of data.
The patent specification makes clear that “signal” is
being used in a narrower sense as a start indicator for the
DRAM to begin an operation. See, e.g., ’109 Patent Ab-
RAMBUS INC. v. REA 9
stract (“A signal is provided that indicates when the
memory device is to begin sampling write data . . . .”).
This court does not need to re-construe the term “sig-
nal,” however. Nor does this court need to reach the
question of whether “signal” encompasses just the “strobe
signal” in the strobe embodiment of the ’109 Patent, or
whether it also encompasses the delay value in the non-
strobe embodiment. Even under the Board’s broad con-
struction, Farmwald does not anticipate the ’109 Patent’s
claims.
The Board held that Farmwald’s method of storing
and recalling delay values from the DRAM’s access-time
registers anticipated the “providing a signal” limitation of
claim 1 in one of four ways. The Board stated,
“Farmwald’s delay value . . . is either “1) in the request
packet; 2) stored in an access-[time register]; 3) provided
for comparison to a clock; and/or 4) implicitly used to
generate another implicit signal after comparison to
signify a match.” J.A. 13. These theories are discussed in
turn below. The court notes that the Board’s failure to
pick one theory of anticipation it found persuasive, or
even most persuasive, itself causes questions about the
Board’s confidence in its anticipation rejection.
1. In the Request Packet
The Board stated that Farmwald’s delay value is “in
the request packet” transmitted from the memory control-
ler to the DRAM. J.A. 13. The record does not support
this view. The record shows that Farmwald’s delay value
is previously stored in an access-time register within the
DRAM. Farmwald col. 6 ll. 40–47; J.A. 10993 (“Farmwald
relies upon the previously received [delay] value sig-
nal . . . .”). Although Farmwald’s request packets contain
a bit that selects an access-time register, this feature
differs from claim 1, which requires “a signal . . . [that]
indicates when the memory device is to begin sampling
write data.” ’109 Patent cl. 1. The bit in Farmwald’s
10 RAMBUS INC. v. REA
request packet does not contain any timing information,
but merely selects an access-time register. Although
subsequent events or signals from the access-time regis-
ters may cause the DRAM to begin an operation, that part
of the Farmwald operation does not transform the bit into
a signal that meets the limitations of the claim. The
invention claimed in the ’109 Patent eliminates the need
for access-time registers—a significant advantage over
the prior art. ’109 Patent col. 10 ll. 27–39.
2. Stored in an Access-Time Register
The Board alternatively held that “storing the delay
value in the access-time register” in Farmwald anticipates
the “providing a signal” step of claim 1. J.A. 14. Howev-
er, Farmwald’s delay values are stored in the access-time
registers prior to the memory controller transmitting any
request packets to the DRAM. Thus, the Board’s inter-
pretation requires performance of the first two steps of
claim 1 out of order. While the steps of a method are not
limited to a specific order unless the claim explicitly or
implicitly so requires, Baldwin Graphics Sys., Inc. v.
Siebert, Inc., 512 F.3d 1338, 1345 (Fed. Cir. 2008), here, a
plain reading of claim 1 evinces a specific order. The first
step of claim 1 “specifies that a write operation be initiat-
ed in the memory device,” and the second step “indicates
when the memory device is to begin sampling write
data.” ’109 Patent cl. 1. It would make no sense for the
second step to be performed first—telling the memory
device to begin sampling write data—before the memory
device was even instructed to perform a “write” operation.
Therefore, the Board’s finding is clearly erroneous.
The Board also stated that “skilled artisans would
have considered . . . retrieving [the delay value]” from
Farmwald’s access-time register as constituting “provid-
ing a signal.” J.A. 14 (emphasis added). The record also
supplies no support for this finding. Farmwald’s access-
time registers are within the DRAM, and thus, the DRAM
RAMBUS INC. v. REA 11
is essentially “retrieving” the delay value from itself.
Claim 1 of the ’109 Patent requires “[p]roviding a signal to
the memory device.” ’109 Patent cl. 1 (emphasis added).
The Board’s reasoning that “[c]laim 1 does not preclude
providing the signal from one part of the memory device
to another part of it,” J.A. 15 n.3, can neither be recon-
ciled with the plain language of the claim nor the patent
specification, which is replete with references and figures
showing an external memory controller transmitting
signals to the DRAM. See, e.g., ’109 Patent col. 4 ll. 8–26.
3. Provided for Comparison to a Clock
The Board stated that “skilled artisans would have
considered . . . comparing [Farmwald’s delay value] to a
clock value . . . as also constituting ‘providing a signal,’ as
recited in claim 1.” J.A. 14. However, nowhere did the
Board identify where Farmwald discloses this compari-
son; nor did the Board explain how “comparing” a delay
value to a clock value equates to a “signal,” particularly
under the Board’s construction of “signal” to mean “the
physical representation of data.” This court has made
clear that “[b]road conclusory statements standing alone
are not evidence.” In re Kotzab, 217 F.3d 1365, 1370 (Fed.
Cir. 2000). In this case, the Board does not show that
comparing a delay value to a clock signal anticipates the
claim.
4. Implicitly Used to Generate Another Implicit Signal
After Comparison to Signify a Match
The Board again, without any discussion or citation to
Farmwald, determined that “Farmwald’s system implicit-
ly compares a clock signal to [the] register-stored delay
value, and thereafter implicitly generates another signal
for completing the data read or write operation upon a
determination that the clock counter equals the delay
value. . . .” J.A. 12. The record does not support this
finding. Furthermore, the implicit signal relied on by the
Board appears to be generated from the access-time
12 RAMBUS INC. v. REA
registers. As mentioned above, Farmwald’s access-time
registers are located within the DRAM itself, and claim 1
of the ’109 Patent requires a signal be provided to the
DRAM from an external source. Therefore, even assum-
ing that Farmwald’s system implicitly generates a signal
upon determination that the clock counter equals a delay
value, the signal would not anticipate the claim.
VI.
In conclusion, the Board’s determination that all 25
claims of the ’109 Patent are invalid as anticipated by
Farmwald is not supported by substantial evidence.
Accordingly, this court reverses.
REVERSED
United States Court of Appeals
for the Federal Circuit
______________________
RAMBUS INC.,
Appellant,
v.
Teresa Stanek Rea, ACTING DIRECTOR, UNITED
STATES PATENT AND TRADEMARK OFFICE,
Appellee.
______________________
2012-1480
(Reexamination No. 95/001,166)
______________________
Appeal from the United States Patent and Trademark
Office, Board of Patent Appeals and Interferences.
______________________
WALLACH, Circuit Judge, dissenting.
The Board’s conclusion that the patent is anticipated
is supported by substantial evidence, and therefore I
dissent from the majority’s decision.
The dispositive question in this case is whether sub-
stantial evidence supports the Board’s finding that
Farmwald discloses the step of “providing a signal to the
memory device, wherein the signal indicates when the
memory device is to begin sampling write data. . . .” ’109
patent col. 42 ll. 61–63. It is uncontested that Farmwald’s
DRAM contains “access-time registers” with various delay
values. Farmwald teaches sending a “request packet”
from a memory controller to the DRAM with control
2 RAMBUS INC. v. REA
information that, inter alia, selects an access-time regis-
ter. The Board found that this request packet was a
“signal” that “indicates” when the DRAM “is to begin
sampling write data.” This is supported by substantial
evidence. Although the delay value choices are stored in
the DRAM, a reasonable person could conclude that the
request packet is a “signal” that “indicates” when the
DRAM “is to begin sampling write data.” Id. col. 42 ll. 61-
63; see Consol. Edison Co. v. Nat’l Labor Relations Bd.,
305 U.S. 197, 229 (1938) (Substantial evidence “means
such relevant evidence as a reasonable mind might accept
as adequate to support a conclusion.”); In re Kotzab, 217
F.3d 1365, 1369 (Fed. Cir. 2000) (“Substantial evidence is
something less than the weight of the evidence but more
than a mere scintilla of evidence.”).
The majority says that although Farmwald’s request
packets contain a bit, that bit merely selects an access-
time register; it does not contain any timing information
itself. However, given the Board’s broad construction of
the term “signal,” which the majority accepts, see Maj. Op.
9, the bit in the request packet provides a signal by select-
ing an access-time register which indicates when the
DRAM is to begin an operation. It is the bit that selects
an access-time register, and therefore indicates “when the
memory device is to begin sampling write data.” ’109
patent col. 42 ll. 62–63. That is, the request packet in-
cludes a bit, a “value,” that tells the memory device to
wait x amount of time after receiving the command to
write data. The fact that the x is stored within the
memory device does not change what the value repre-
sents. Without the value in the request packet the
memory device would not know when to begin sampling;
it is that bit which instructs the memory device when to
begin. Farmwald teaches that the request packet controls
the timing of writing data either by directly “select[ing] a
certain register in the slave DRAM memory device which
stores the (delay value) timing information” or indirectly
RAMBUS INC. v. REA 3
by “indicat[ing] pre-selected (delay value) access times.”
See J.A. 8 (citing Farmwald col. 9 l. 46 – col. 10 l. 5)).
Contrary to the majority’s characterization, no subse-
quent steps or signals are necessary. See Maj. Op. 8–9.
Once the specified amount of time elapses, sampling
begins. Farmwald col. 27 ll. 19–20 (Claim 25 specifies
“outputting the data to the memory device after a delay
time transpires.”).
The second basis for the Board’s reasoning is that the
claim language does not require sequential reading of the
first two steps. The first two steps state:
Providing control information to the memory de-
vice, wherein the control information includes a
first code which specifies that a write operation be
initiated in the memory device;
Providing a signal to the memory device, wherein
the signal indicates when the memory device is to
begin sampling write data, wherein the write data
is stored in the memory core during the write op-
eration . . . .
’109 patent col. 41 l. 64 – col. 42 l. 64. The “write opera-
tion” instruction does not need to come before the claimed
“signal” since that signal only indicates when the “write
operation” is to happen. There is no reason the write
instruction and the signal indicating the timing of the
write operation could not come at the same time—like
they do in Farmwald’s request packet.
The majority states that “a plain reading of claim 1
evinces a specific order,” because “[i]t would make no
sense for the second step to be performed first—telling the
memory device to begin sampling write data—before the
memory device was even instructed to perform a ‘write’
operation.” Maj. Op. 10. However, it is not clear, as a
matter of “logic or grammar,” that all of the steps in claim
1 must be performed in the order written. See Altiris, Inc.
4 RAMBUS INC. v. REA
v. Symantec Corp., 318 F.3d 1363, 1369–71 (Fed. Cir.
2003). “[A]lthough a method claim necessarily recites the
steps of the method in a particular order, as a general
rule the claim is not limited to performance of the steps in
the order recited, unless the claim explicitly or implicitly
requires a specific order.” Baldwin Graphic Sys., Inc. v.
Siebert, Inc., 512 F.3d 1338, 1345 (Fed. Cir. 2008). The
third step, “providing a first bit of the write data to the
memory device during an even phase of a clock signal,”
and the fourth step, “providing a second bit of the write
data to the memory device during an odd phase of the
clock signal,” have parallel structure and sequential
language. ’109 patent col. 42 ll. 66–67; col. 43 ll. 1–2. But
the same cannot be said for the first two steps of claim 1.
The first step describes a “first code” but the claim fails to
identify a “second code.” There is no sequential language
in the first two steps to indicate that they must occur in
the order written. The broadest reasonable interpretation
of signal indicates when the operation is to happen, but
does not necessarily require immediate initiation. There-
fore, in instances like the ’109 patent’s delay value embod-
iment, wherein the signal instructs the DRAM to wait
some number of clock cycles before initiating action, it
does not matter whether the DRAM has been instructed
to do a write operation until the specified period of time
elapses.
The role of this court is not to determine if the record
could support a different outcome, but to determine if the
Board’s findings are supported by substantial evidence.
In re Jolley, 308 F.3d 1317, 1320 (Fed. Cir. 2002) (If “the
evidence in [the] record will support several reasonable
but contradictory conclusions,” then this court “will not
find the Board’s decision unsupported by substantial
evidence simply because the Board chose one conclusion
over another plausible alternative.”). Because the Board’s
findings on anticipation are supported by substantial
evidence, I respectfully dissent.