Full opinion text
OPINION AND ORDER FOLLOWING MARKMAN HEARING McMAHON, District Judge. Plaintiffs TM Patents, a Delaware limited partnership formed under the Plan of Reorganization of Thinking Machines Corporation (“Thinking Machines”) for the purpose of recovering for infringement of the corporation’s patents, and TM Creditors, a Delaware limited liability corporation that represents the interests of the bankrupt’s unsecured creditors, are shared successors-in-interest to the patent estate of Thinking Machines, a developer and manufacturer of computers and computer peripheral equipment. Plaintiffs commenced this action against defendant IBM, alleging that certain IBM products infringed three of TM’s patents: Patent No. 5,202,979 (the ’979 patent); Patent No. 4,899,342 (the ’342 patent); and Patent No. 5,212,773 (the ’773 patent). The first two patents relate to technologies for storage of data in computers and the detection and correction of errors in such data; the third deals with a strategy for routing communications in massively parallel processors. The matter is before the Court for construction of the three patents in suit, as required by Markman v. Westview Instruments, Inc., 517 U.S. 370, 116 S.Ct. 1384, 134 L.Ed.2d 577 (1996). After extensive briefing, the parties presented their respective contentions concerning patent construction in light of intrinsic evidence to the Court at a hearing conducted September 8 and 9, 1999. The hearing was attended by, inter alia, Professor Jack Lipovski, who had previously been appointed as the Court’s disinterested expert and technical advisor. (See Order dated July 21, 1999.) The September hearing was limited to issues of intrinsic evidence because, in determining the meaning of the patent claims, the Court must first examine such evidence — including the claims themselves, the specification, and the prosecution history — and determine whether it can derive an unambiguous construction of the claim language from those sources. See Vitronics Corp. v. Conceptronic, Inc., 90 F.3d 1576, 1582 (Fed.Cir.1996). As Judge Conner of this Court recently observed, “If this intrinsic evidence permits an unambiguous construction of the claim language, the Court need not, and indeed should not, consider extrinsic evidence, such as testimony from expert witnesses as to special meanings which the terms of the claims have for those skilled in the art.” Evans Medical Ltd. v. American Cyanamid Company, 11 F.Supp.2d 338, 347 (S.D.N.Y.1998) (citing Vitronics, 90 F.3d at 1585), aff'd, Nos. 98-1440, 98-1459, 1999 WL 594310 (Fed.Cir. Aug. 9, 1999). Extrinsic evidence should only be relied upon where necessary to resolve an ambiguity in a disputed claim term. See CVI/Beta Ventures, Inc. v. Tura, L.P., 112 F.3d 1146, 1153 (Fed.Cir.1997). Finding the intrinsic evidence sufficient for Markman purposes, the Court construes the patents as set forth below. A The ’342 Patent The first patent in suit is a patent for a computer system that has as one of its properties the ability both to detect and to correct errors in data that is stored in the computer. The salient features of the system are its use of an error correction code (“ECC”) that has the capacity not just to detect errors in data but to correct them as well, and spare disk drives that “back up” the corrected data, so that there are always two copies in the system. The disk storage array described in the specifications for the ’342 patent consists of 32 dedicated data disks and 7 dedicated ECC disks. In addition, there are 3 spare disks in the array. The disk storage array is connected to a computer by a 64-bit bus, or conductor, along which data is transmitted. The bus is in turn connected to a specialized bus adapter, including an ECC generation circuit, within the array. The bus adapter is thus connected to the disk drive. During transmission of data to the disk array, a 64-bit wide data stream enters the adapter and is split into two 32-bit groups. Each of these 32-bit groups is transmitted to the ECC generation chip to generate a 7-bit ECC. Thereafter, each block of 32 bits and the accompanying ECC bits are sent to all 39 disks in the array through 39 shift registers (computer hardware elements designed to perform shifting of the data contained within them). Data is stored in what is referred to as “bit-striped mode,” meaning that one bit of each of the 32-bit groups of data is stored separately on each of the 32 disk drives. The 32 data bits and ECC bits are sent together from all 39 disks in the array via the shift registers to a specialized bus adapter, where, using the data and error correction code, the ECC generation circuit chip performs an error detection and correction function on the data bits and delivers the corrected data to the computer. If there is a disk drive failure, the corrected data can be accessed from the spare disk. The claims in the ’342 patent that require construction are Claims 1, 6, 7 and 10. 1. Collateral Estoppel Effects from the EMC Action TM’s suit against IBM is not the first infringement action to reach the stage of a Markman hearing in a federal District Court. In 1997, TM also sued a competitor of IBM, EMC Corporation, in a case heard by Chief Judge William G. Young in the United States District Court for the District of Massachusetts. See TM Patents v.. EMC Corp., Civil Action No. 98-10206 (D.Mass. Jan. 27, 1999). The case went to trial earlier this year. Immediately prior to the trial, Chief Judge Young held a Markman hearing, at which he was asked to construe some, but not all, of the claims disputed in this Court. Judge Young did so, after a two day hearing. IBM asserts that TM is either collaterally or judicially estopped to relitigate the claims that Judge Young construed — with which constructions IBM (a non-party to the EMC action) is in substantial (though not total) agreement. TM argues that Judge Young’s rulings, while correct in many respects and perhaps persuasive in others, ought not be accorded former adjudication effect, because the EMC action settled during trial, and because Judge Young’s rulings were not sufficiently “final” to be deemed preclusive. IBM is correct. While this raises an issue of first impression, I conclude that Judge Young’s resolution of the meaning of certain disputed patent terms following a Markman hearing, at which TM had a full and fair opportunity to litigate the meaning of those terms, is binding on the Plaintiffs in this action. Collateral estoppel forecloses litigants from contesting matters that were actually litigated and decided in a previous action. See Blonder-Tongue Lab. Inc. v. University of Illinois Foundation, 402 U.S. 313, 329, 91 S.Ct. 1434, 28 L.Ed.2d 788 (1971). Four elements must be met for collateral estoppel to apply. First, the issues raised in both proceedings must be identical. Second, the relevant issues must have actually been litigated and decided in the prior proceeding. Third, the party to be estopped must have had a full and fair opportunity to litigate the issues in that prior proceeding. And fourth, resolution-of the issues must have been necessary to support a valid and final judgment on the merits. See Central Hudson Gas & Elec. Corp. v. Empresa Naviera Santa, 56 F.3d 359, 368 (2d Cir.1995). The parties before me agree that the first and third of those four elements are met in this case. They disagree, howevér, as to whether the second and fourth have been satisfied. Since in this particular case, the fourth element (finality) subsumes the second (actually decided), I will address them together. As to the fourth element, TM notes that the dispute between it and EMC concerning the meaning of certain terms in the ’342 patent was never reduced to a final judgment, because the matter was settled before the jury had returned its verdict on the question of infringement. TM adopts the simple and straightforward position that no final, appealable judgment means no finality for collateral estoppel purposes. Unfortunately for TM, that is not the law in this Circuit (or any other, for that matter). Since Judge Friendly’s seminal opinion in Lummus Co. v. Commonwealth Oil Ref. Co., 297 F.2d 80, 89 (2d Cir.1961), it has been settled that a judgment that is not “final” in the sense of 28 U.S.C. § 1291 can nonetheless be considered “final” in the sense of precluding further litigation of issues that were actually determined in such a judgment. Whether a ruling is sufficiently final turns on “such factors as the nature of the decision (i.e., that it was not avowedly tentative), the adequacy of the hearing, and the opportunity for review.” See id. As Judge Friendly observed, “ ‘Finality’ in the context here relevant may mean little more than that the litigation of a particular issue has reached such a stage that a court sees no really good reason for permitting it to be litigated again.” Id.; see also Zdanok v. Glidden Co., Durkee Famous Foods Div., 327 F.2d 944, 955 (2d Cir.1964); Metromedia Co. v. Fugazy, 983 F.2d 350, 366 (2d Cir.1992). The Lummus principle is illustrated by Georgakis v. Eastern Air Lines, Inc., 512 F.Supp. 330 (E.D.N.Y.1981). In Georgak-is, the plaintiff, who was injured in an airplane crash while traveling with the defendant airline, sought summary judgment on the airline’s affirmative defense that the case was covered by the Warsaw Convention, which would have limited the plaintiffs damages to $75,000. In a prior case arising out of the same crash brought by a different passenger, the court had granted summary judgment against the airline, dismissing the Warsaw Convention defense. The passenger in Georgakis invoked collateral estoppel. Rejecting the airline’s argument that the interlocutory nature of the earlier decision precluded application of collateral estoppel, the court noted that collateral estoppel “does not require a judgment ‘which ends the litigation ... and leaves nothing for the court to do but execute the judgment,’ ... but includes many dispositions which, although not final in that sense, have nevertheless been fully litigated.” Id. at 334 (quoting Zdanok, 327 F.2d at 955). The parties have not called my attention to any case in which a court has applied collateral estoppel to bar relitigation of claims construction issues decided at a pri- or Markman hearing, and I have not located any such decisions. Nonetheless, that proposition seems self-evident. Markman ushered in a new regime in patent claims construction. Prior to the Federal Circuit’s ruling in 1995, disputes concerning the meaning of patent claims were submitted to a jury along with questions about validity and infringement. Thus, until there was a final judgment after a jury verdict, there was no construction of claims, see, e.g., Tol-O-Matic, Inc. v. Proma Produkt-Und-Marketing Gesellschaft, m.b.H., 945 F.2d 1546 (Fed.Cir.1991), and hence, no finality for collateral estoppel purposes. However, after Mark-man, claim construction became a separate legal issue, for determination by the Court. The parties frequently litigate the meaning of those limitations prior to the trial, as occurred in both EMC and this case, so that the Court can instruct the jury on the meaning of the patent at the outset of the case. Moreover, the Court limits itself to construing that which is necessary to the resolution of the questions of infringement and validity. The jury is not free to override the Court’s construction of the disputed terms. It is hard to see how much more “final” a determination can be. The facts of the EMC case bear out this interpretation. The Massachusetts court was required to construe any disputed terms in the ’342 patent for the jury. Pri- or to the trial in EMC, Judge Young held a two day hearing, at which both parties were ably represented — TM by the very same lawyers who represent it in the instant action. The parties identified certain claim limitations on whose meaning they could not agree, and Judge Young heard whatever evidence he thought necessary to interpret those limitations. He then issued a very thorough ruling, disposing of all disputed issues. After ruling, he entertained reargument and made several modifications. He then read his ruling to the jury on the first day of trial, in a preliminary jury instruction. Judge Young used the claim construction to guide his eviden-tiary rulings. He also gave the jurors copies of what he had read, so that they could refer to his construction during the course of the trial. The jurors were not free to adopt a contrary construction of the patent claims in suit, and they were so advised. A verdict would not have changed anything about Judge Young’s Markman rulings. Nothing more remained to be adjudicated; nothing more remained to be decided on the issue of claim construction. The application of the claim to the product was immaterial to the finality of Judge Young’s determinations. Thus, under Lummus and its progeny, the results of the Markman hearing in the EMC action were sufficiently “final” to permit application of collateral estoppel — even though the matter to which they were necessary was never reduced to a final judgment after verdict. See also Restatement (Second) of Judgments § 13, comment e (1980) (“A judgment may be final in a res judicata sense as to part of an action although the litigation continues as to the rest.”); Sherman v. Jacobson, 247 F.Supp. 261, 268 (S.D.N.Y.1965) (noting that a judgment “may be final as to some matters, even though the litigation continues as to others”). It bears noting that one of the Supreme Court’s rationales for upholding the Federal circuit’s ground-breaking decision in Markman was the promotion of uniformity in the meaning to be given to a patent claim. See Markman, 517 U.S. at 390-91, 116 S.Ct. 1384. Even prior to Markman, the Federal Circuit had held that determination of the scope of a patent claim in a prior infringement action could have collateral estoppel effect against the patentee in a subsequent case. See Pfaff v. Wells Elec. Inc., 5 F.3d 514, 517-18 (1993). After Markman, with its requirement that the Court construe the patent for the jury as a matter of law, it is inconceivable that a fully-litigated determination after a first Markman hearing would not be preclusive in subsequent actions involving the same disputed claims under the same patent. The nature of the Markman proceeding is such that finality is its aim. TM contends that Judge Young’s determinations were “avowedly tentative” for several reasons. None is persuasive. First, Judge Young expressed the view that he was “shaky” about several of his conclusions, and indicated that they were “subject to revision.” (Joint Appendix on the Issues of Collateral Estoppel and Claim Construction at A 14-15.) Of course, a Court does not vitiate the collateral estoppel effect of a ruling by expressing some doubt about its conclusions. See Restatement (Second) of Judgments § 13, comment g (“The test of finality ... is whether the conclusion in question is procedurally definite and not whether the court might have had doubts in reaching the decision”); United States v. McGann, 951 F.Supp. 372, 380-81 (E.D.N.Y.1997). However, Judge Young made these statements while inviting the parties to move for reconsideration if they disagreed with his constructions (a procedure this Court is not inclined to emulate). The parties accepted his invitation — in particular, TM asked for reconsideration of several issues that all parties believe are key to the question of infringement. Notably, Judge Young ceased to feel “shaky” after further briefing and argument, telling the parties, “You’ve got the claim construction, you’re going to have to live with it.” (A 23-23.1.) Second, TM has cited several passages from the trial transcript in support of its argument that Judge Young had only gone “so far” in construing certain phrases. In particular, TM argues that the Judge had not defined the term “integrated” as used in the phrase “single integrated circuit,” (A 42-44), and contends that this constitutes evidence that Judge Young’s rulings were not intended to be final. However, at most this omission would mean that Judge Young cannot bind me to a definition of the term “integrated” as used in the phrase “single integrated circuit.” It does not mean that his other rulings were avowedly tentative, especially when both his words and his behavior during the trial are to the contrary. Plaintiffs next argue that the Markman constructions were never reviewed by a higher Court — a not insignificant proposition when nearly 40 percent of claims constructions are changed or overturned by the Federal Circuit. See Cybor Corp. v. FAS Technologies., Inc., 138 F.3d 1448, 1476 (Fed.Cir.1998) (Rader, J., dissenting). However, the only reason Judge Young’s conclusions were not reviewed on appeal is that the case was settled. A party who cuts off his right to review by settling a disputed matter cannot complain that the question was never reviewed on appeal. The Markman rulings were not vacated as part of the settlement. They therefore remain preclusive. See Hartley v. Mentor Corp., 869 F.2d 1469, 1472-73 (Fed.Cir.1989); Wellons, Inc. v. T.E. Ibberson Co., 869 F.2d 1166, 1169 (8th Cir.1989); Siemens Med. Sys., Inc. v. Nuclear Cardiology Sys., Inc. 945 F.Supp. 1421, 1436 (D.Colo.1996); Ossman v. Diana Corp., 825 F.Supp. 870, 878 (D.Minn.1993). Fourth, TM argues that a Markman adjudication cannot be preclusive prior to verdict because there is no requirement that a Court construe the claims in suit prior to giving its final, instruction to the jury. Indeed, it appears to be the practice in at least one federal district court to hold the Markman hearing during the trial, with the Court issuing its ruling when it instructs the jury at the close of the case. See, e.g., Johns Hopkins Univ. v. Cellpro, 894 F.Supp. 819, 826 (D.Del.1995). Of course, if Judge Young had followed such a procedure, then there would be no issue preclusion from the EMC case. But he did not. He held a two day hearing, following which he issued a ruling construing the claims. He then heard reargument and fine-tuned his ruling, telling the parties to “live with it.” He read the ruling to the jury and handed each juror a copy. The fact that he could have done otherwise does not vitiate the preclusive effects of what he actually did. Had Judge Young followed the Delaware procedure and announced his Markman ruling at the close of the evidence, only to see the case settle while the jury was deliberating, I have no doubt that his views about the meaning of the disputed claims would be just as pre-clusive as if they had been announced pri- or to trial. The fact that the judge makes the ruling is sufficient under Lummus; when he does so is irrelevant. The cases TM cites for the proposition that the collateral estoppel effects from claim construction should be narrowly limited to matters that were essential to a judgment (of validity or infringement)— cases such as Jackson Jordan, Inc. v. Plasser Amer. Corp., 747 F.2d 1567 (Fed.Cir.1984); A.B. Dick Co. v. Burroughs Corp., 713 F.2d 700 (Fed.Cir.1983); and Studiengesellschaft Kohle v. Eastman Kodak Co., 616 F.2d 1315 (5th Cir.1980) — are inapplicable in the post-Markman era, at least when the district court holds a special pre-trial hearing, as Chief Judge Young did in the EMC case. These authorities were decided at a time when patent claims were construed during jury deliberations. It therefore made perfect sense to limit the collateral estoppel effects emanating from a jury’s judgment on issues of validity and infringement to matters of claim construction that were necessarily comprehended in the verdict. That rule makes no sense when a court, acting as a matter of law, draws binding conclusions about the meaning of disputed patent terms for the benefits of the litigants and the jurors. Even the one post-Markman case cited by TM for this proposition, Phonometrics, Inc. v. Northern Telecom, Inc., 133 F.3d 1459 (Fed.Cir.1998), is not inconsistent with the principle adopted by this Court. Although the Phonometrics case was decided after Markman, it followed an earlier decision that was not given preclu-sive effect and was decided several years before Markman. See Intellicall, Inc. v. Phonometrics, Inc., 952 F.2d 1384 (Fed.Cir.1992). Finally, TM argues that a recent Federal Circuit case, Odetics, Inc. v. Storage Tech. Corp., 185 F.3d 1259 (Fed.Cir.1999), worked a change in the law and demonstrated that some of Judge Young’s rulings were erroneous. To that extent, TM contends, collateral estoppel ought not be applied, since it would be clear error to give preclusive effect to a ruling that was entered prior to “a significant change in the legal atmosphere.” Bingaman v. Department of the Treasury, 127 F.3d 1431, 1437-38 (Fed.Cir.1997), citing Commissioner v. Sunnen, 333 U.S. 591, 600, 68 S.Ct. 715, 92 L.Ed. 898 (1948). TM is correct about the legal principle, but incorrect when it asserts that Odetics worked a change in the law. The Odetics opinion cited above was the eighth and most recent in the protracted Odetics litigation. In that decision, the Federal Circuit did not address the issue of claim construction, as TM asserts, but rather reviewed the trial court’s infringement decision, notably, its comparison of the accused product with the previously construed patent claim (construed under the means-plus-funetion limitation). The claim limitation was construed in an earlier decision, Odetics, Inc. v. Storage Tech. Corp., 1997 WL 357598 (Fed.Cir. Jun. 25, 1997). Thus, the recent Odetics ruling did not work any change in the law with respect to that issue, although the decision will prove relevant to some of the issues of construction on claims that are not subject to collateral estoppel. If “finality” in the collateral estoppel context “mean[s] little more than that the litigation of a particular issue has reached such a stage that the court sees no really good reason for permitting it to be litigated again,” Lummus, 297 F.2d at 89, I see no reason to permit TM a second chance to litigate the meaning of the claims previously construed by Judge Young. Therefore, I hold that collateral estoppel forecloses TM from relitigating the meaning of limitations (ii), (iii), (vii) and (viii) of Claim 1, and the corresponding limitations (i), (iii) and (iv) of Claim 7, see infra at part 2(c). The other claims in dispute before me were not disputed in Massachusetts, and TM is free to argue for any interpretation of those limitations here. And, of course, IBM is not precluded from litigating anything at all. So there is plenty for this Court to decide from scratch. Finally, I have to observe that this issue of collateral estoppel, which has taken on a life of its own in this action (no doubt due to my preoccupation with it), is of marginal practical importance, because I agree with just about everything Judge Young did when he construed the claims in the EMC action. Nonetheless, I have no doubt that collateral estoppel would apply against TM on the previously litigated claims even if I thought everything Judge Young decided was wrong. This does not, however, mean that I agree with all the “gloss” that IBM puts on certain key rulings by Judge Young, a point which I discuss further below. IBM has relied upon two other issue preclusion theories, judicial estoppel and party admissions under Fed.R.Evid. 801(d). It is not necessary for me to reach those arguments, although I do note that IBM’s invocation of judicial estoppel is improper, as that doctrine applies only to inconsistent factual positions, not to issues of law. See Bates v. Long Island RR Co., 997 F.2d 1028, 1037 (2d Cir.1993); TLC Beatrice Int’l Holdings, Inc. v. CIGNA Ins. Co., 1999 WL 33454 at *7 (S.D.N.Y. Jan.27, 1999). 2. Construction of the Disputed Claims in the ’812 Patent Before going to the individual disputed claims, I observe that the ’342 patent does not seem to me a particularly difficult patent to construe. For the most part, the words are clear and unambiguous and can be given their plain meaning. My reading of Judge Young’s decision is that he did precisely that, and I have no fundamental disagreement with any of his conclusions. Most of what IBM asks me to overlay onto Judge Young’s prior construction of Claim 1 (which is where the disputed issues arise) derives from the patent specification. IBM takes the position that the only invention disclosed in the patent application was the preferred embodiment, in all its detail. However, IBM overlooks settled rules of patent construction that are discussed below. Absent an indication that the plain terms of the patent were not intended to control, I decline to follow IBM’s suggested approach to interpretation of the claims. (а) Claim No. 1 (Independent Claim): Disputed Limitations (i) 1. A multi-unit memory system comprising: This claim, which was not disputed before Judge Young, means precisely what the words say. “A multi-unit memory system” is “a” (that is to say, one, or a single) system that uses two or more data storage memory units. Of course, the fact that the patent claims “a” system does not mean that IBM or some other party would escape liability for infringement by constructing two or three or even more such multi-unit memory systems and somehow linking them together or causing them to operate together. There is no reason to read anything further into the very simple words used by the claimant. There is no mention of a “single array” in the patent claim, and I decline to adopt IBM’s suggestion that I read such a limitation into the claim, applying the settled principle of patent construction that the language of the claim defines its scope, see Mantech Environmental Corp. v. Hudson Environmental Servs., Inc., 152 F.3d 1368, 1373 (Fed.Cir.1998), and that the words of a claim are to be construed in accordance with their ordinary meaning to persons in the relevant field of technology, unless it appears that the inventor used them otherwise. See Bell Communications Research, Inc., v. Vitalink Communications Corp., 55 F.3d 615, 620 (Fed.Cir.1995). In particular, claims are not to be limited or construed by reading in extraneous words from written descriptions or specifications that are not contained in the claims themselves. See Laitram Corp. v. NEC Corp., 163 F.3d 1342, 1347 (Fed.Cir.1998). This rule is not inconsistent with the principle that claims should be construed in light of the patent specification, see, e.g., Bell Communications, 55 F.3d at 620, but simply recognizes that the claims of a patent are not limited to the preferred embodiment that is described in the specifications. See Ekchian v. Home Depot, Inc., 104 F.3d 1299, 1303 (Fed.Cir.1997). Throughout this proceeding, IBM has contorted the plain meaning of words and phrases used in the patent claims themselves by referring to the written description, including the specifications and the embodiment disclosed therein. In construing this patent (and, indeed, all three patents), I have adopted the principle that, where the language of the claim is clear and unambiguous, I will read nothing additional into it. If this means that the patent turns out to be invalid (which is a decision for the jury, not for the Court), then so be it. Although courts are to construe claims so as to sustain a patent’s validity where possible, see ACS Hosp. Systems, Inc. v. Montefiore Hosp., 732 F.2d 1572, 1577 (Fed.Cir.1984), courts are nonetheless bound to follow the other rules of construction discussed above. See id. (ii) ... an error correction code generation circuit which generates for each block of data to be stored in said memory system an error correction code, ... (iii) ... each block of data comprising a first plurality of digits and each error correcting code comprising a second plurality of digits from which at least one digit of error may be detected and corrected in the block of data ... This is the first claim that is affected by collateral estoppel from Judge Young’s decision. In the EMC case, Judge Young construed this claim language as follows: “[Error correction code generation circuit] means a single integrated circuit which generates the claimed error correction code ... Data, as I have already said, is information. Here, the data is in the form of digits. A block of data, as used in this phrase, means a group of data bits which may be corrected by another group of bits called error correction bits. The data is information to be stored in the computer.” (A 17.) This construction, relying as it does on the plain meaning of the words used in the claims limitation, is plainly correct. The only arguably unclear term used by Judge Young is his definition for the phrase “error correction code circuit,” which he defined as a “single integrated circuit which generated the claimed error correction code.” (A 17.) The words “single integrated circuit” do not appear in the claim, and TM urges that Judge Young erred in so limiting its claim. Putting to one side the fact that TM is collaterally estopped to challenge Judge Young’s ruling, TM is simply incorrect. The limitation refers to “an” — i.e., one— error correction code circuit. In other words, the literal language of the claim calls for one circuit that generates an error correction code. The patent specification discloses that this circuit was a single integrated circuit — that is, “... an interconnected array of active and passive elements integrated with a -single semiconductor substrate or deposited on the substrate by a continuous series of compatible processes, and capable of performing at least one complete electronic circuit function.” McGraw Hill Dictionary of Scientific and Technical Terms (3d. ed.1984). Judge Young did not err when he restricted the claim in limitation (ii) to error correction code circuits that generated the correcting digit via a single integrated circuit. That said, I can think of ways to phrase parts of the interpretation that might be easier for a jury to understand, and I do not understand the doctrine of collateral estoppel to require me to quote Judge Young verbatim. I will, therefore, rephrase it as follows: The multi-unit memory system I have just described contains a single integrated circuit that is called the error correction code circuit. What does this circuit do? It generates — that is, it creates— something called an error correction code, which is a group of mathematically interrelated bits. It creates one of these error correction codes for each block of data that is to be stored in the memory system. Now, let’s take a look at what some of those words mean — the words that might be unfamiliar to you. Data is simply information. In a computer, data is stored as digits — specifically, as the digits 0 and 1, as will be explained to you during the course of the trial. Each individual digit is known as a bit, which is a contraction of the phrase “binary digit.” For those of you who remember the binary number system from your high school math class — that’s also known as the Base 2 system — 0 and 1 are the two digits in the binary system. So a block of data is two or more data bits that are used to code information in a computer. That’s what the phrase “plurality of digits” means — it’s just a fancy way of saying more than one digit. An error correction code is also a plurality of digits — remember, that means more than one of those binary digits, 0 and 1 — and the error correction code has a particular purpose, which is to detect and correct an error in a block of data. So for every block of data in our multi-unit memory system, there is an associated error correction code, which is capable of doing two things: determining that at least one of the bits in the associated data block is correct, and correcting the error that it has detected, by generating a new, corrected bit of data. IBM, relying on the specification, has urged me to insert a further limitation: namely, that any block of data can contain no more than 32 bits. However, there is no reason to read any such requirement into the claim, since there is nothing about the phrase “block of data” that necessarily suggests the size of the block — as opposed to the phrase “an error code generation circuit” — which does suggest a single circuit rather than multiple circuits. Of course, it is possible that a jury would not find infringement of the Plaintiffs claims by a system that can detect and correct errors in larger blocks of data, but that does not bear on the construction of the claim. (iv) ... a plurality of read/write memory units, which store the digits of said blocks of data and associated error correction codes generated by said error correction code generation circuit and read said digits ... (v) ... at least some of the different digits of each code and its associated data block being stored in different memory units. Again, these words are not difficult to interpret in light of their plain meaning. These limitations require that there be more than one memory unit to store the blocks of data and their associated error correction codes. These units must be what are known as read/write units — that is, they must be able to store the data bits and to read them. Moreover, at least some of the different digits of each code and its associated data block must be stored in different memory units. That is, if we have a data block reading 100101001, and an error correction code that corresponds to that particular data block, at least some of the digits of the data block must be stored in a different memory unit from the error correction code. IBM argues strenuously that no portion of the error correction code can reside in the same memory unit where the underlying block of data is found, relying on the specifications and the preferred embodiment, which clearly show 32 memory units storing only data and 7 associated memory units storing only error correction codes. The claim limitation as written, however, encompasses more than the configuration urged by IBM. This conclusion is reinforced by looking at unasserted Claim 5, which depends on Claim 1 and which clearly states that the all the error correction code bits must be stored on separate memory units from the data block digits. Under the rule of claims differentiation, when some claims are broad and other, dependent claims are explicitly narrower, the limitation of the narrower claim cannot be read into the broader claim, even to avoid invalidity or to escape infringement. See Karlin Technology Inc. v. Surgical Dynamics, Inc., 177 F.3d 968, 971-72 (Fed.Cir.1999); Uniroyal, Inc. v. Rudkin-Wiley Corp., 837 F.2d 1044, 1054-55 (Fed.Cir.1988); D.M.I., Inc. v. Deere & Co., 755 F.2d 1570, 1574 (Fed.Cir.1985). IBM is free to argue at trial that the patent as construed is invalid, but I am not free to overlook the doctrine of claim differentiation in order to construe the claim limitations more narrowly than they are written. Similarly, I reject IBM’s effort to have me rule that claim limitation (iv) requires that the block of data and its associated error correction code be stored across all of the memory units in the array. Again, IBM is trying to convince me that the claim should be construed so as to limit the invention to its preferred embodiment as described in the specifications. I decline to do so. (vi) ... at least one spare read/write memory unit that is similar in operation to an individual read/write memory unit of said plurality of read/write memory units ... Once again, the plain meaning of the words used in the limitation dictates the construction to be given by this Court. The phrase “at least one” does not, as IBM urges, mean “one and no more than one.” Rather, as TM notes, it means “one or more than one.” Therefore, the claim limitation teaches that the multi-unit memory system should have one or more than one “spare” memory unit, with “spare” used in its ordinary dictionary sense (á la “spare tire”). (vii) ... means for generating from the digits of a block of data and associated error correction code read from said memory units a digit which corrects an error in a digit read from one of said memory units, said generating means operating on a sequence of said blocks of data and associated error correction codes to generate a sequence of correct digits, and (viii) ... means for accessing the spare read/write memory unit to store therein said sequence of correct digits as the sequence of correct digits as the sequence of blocks of data and associated error correction codes are read from the plurality of read/write memory units. These two claims must be interpreted in accordance with the dictates of 35 U.S.C. § 112(6), which was inserted into the patent laws after the United States Supreme Court invalidated so-called “means-plus-function” claims limitations. See Halliburton Oil Well Cementing Co. v. Walker, 329 U.S. 1, 67 S.Ct. 6, 91 L.Ed. 3 (1946). Recitations that are properly considered to fall within this rule are construed differently than other limitations, in that any infringing device or process must both perform the identical function as does the claimed invention and must do so with a structure that is identical or equivalent to that described in the specification. See Odetics, supra, 185 F.3d 1259, 1999 WL 455530 at *4. Thus, unlike an ordinary limitation, a means-plus-function limitation is to have the specification read into it. That is precisely what Judge Young did when he construed the above limitations in the EMC litigation. I find no fault with his conclusions — though I do not read them in quite the same way that IBM does. Judge Young construed limitation (vii) as follows: [E]ach means claim requires the description of a function, and then an associated structure to accomplish that function. Here the claimed function is to generate a corrected bit by reading all of the bits of a block of data and its associated error correction code, including any erroneous bit. The disclosed structure of this claim is a Texas Instrument semiconductor chip, or its structural equivalent, which generates a corrected digit using a Hamming code, or its structural equivalent. (A 18.) Both parties seek changes in Judge Young’s reading. TM is, of course, collaterally estopped from challenging Judge Young’s conclusions, but IBM is not. However, the language IBM proposes to add to Judge Young’s reading of the patent is unsupportable. IBM asks this Court to rule that structural equivalent of the Texas Instruments 74AS632 integrated circuit chip is any other integrated circuit chip for generating a Hamming code. The Hamming code is a particular form of error correction code, which both detects and corrects errors by recalculating stored data to match the data when read by the computer. IBM — which uses a different sort of error correction method, known as simple parity, in the alleged infringing products (known as the RAID 3, 4, 5 and 6 systems), insists that any system that detects and corrects data by using any other error correction code necessarily falls outside the ambit of the claimed invention. IBM’s effort to win the battle of infringement at the Markman stage is predictable, but this Court has no intention of usurping the jury’s function. Since even a means-plus-function claim can be infringed by the structural equivalent of a disclosed embodiment, it would be error for me to interpret the claim as being limited to systems that employ Hamming code to detect and correct errors — particularly since a Court is not to construe patent claims with an eye to the product at the Markman phase. See CVI/Beta Ventures, 112 F.3d at 1153. It may be, as IBM urges, that there is no structural equivalent to the Hamming code, and that its RAID 3, 4, 5 and 6 systems do not infringe the patents in this case because they perform the identical function but in a fundamentally different way. That, however, is a question of equivalence. It is not before this Court today. Moreover, IBM’s insistence that such details as Hamming code and Texas Instrument chips are literally part of what the inventor under the ’342 patent is claiming, while understandable, is unwarranted. Here, the recent Odetics case is instructive. In that case, the district court ruled that the individual components of an overall structure that corresponded to a claimed function — in that case, a “rotary means” for accessing stored videocassettes — were part of the claim limitation. Although, as IBM notes, the issue of claim construction was not directly before the court, the Federal Circuit reversed, stating: The claim limitation is the overall structure corresponding to the claimed function. That is why structures with different numbers of parts may still be equivalent under Section 112, para. 6, thereby meeting the claim limitation. The appropriate degree of specificity is provided by the statute itself; the relevant structure is that which “corresponds” to the claimed function. Further deconstruction or parsing is incorrect. Odetics, 1999 WL 455530 at *6. Judge Young was quite careful to rule that structural equivalents of the TI semiconductor chip (i.e., single integrated circuits) and structural equivalents of the Hamming code (which he found to be “other codes that perform precisely the same function, which were known at the time of this patent”) fell within the scope of the claimed invention. Try as it might, IBM cannot convince me to read the word “equivalents” out of the claim construction. I adopt Judge Young’s construction of limitation (vii). The only disputed portion of limitation (viii), the “means for accessing” claim, is Judge Young’s description of the disclosed structure. Otherwise, both parties are agreed that this limitation requires the ability to access the “spare” memory/memories for the purpose of storing a “back-up” set of corrected digits at the same time that the corrections are being sent to the main memory/memories. I can see no reason for adding IBM’s proposed language to Judge Young’s carefully crafted description of the disclosed structure — ■ “a set of multiplexers, shift registers, and control signals that direct a corrected bit to the spare disk drive or its equivalent ... ” — and I decline to do so. IBM’s proposal to have a claim construction that “fixes the location of the spare memory unit in the single array of memory units” is grounded, like so many of its arguments, in its desire to have the Court limit the claims to the details of the specification— an approach that I have rejected. IBM’s proposal also runs afoul of my decision that the phrase “at least one spare memory unit” is not restricted to the singular. (b) Claim 6 (Dependent Claim): Disputed Limitations There are no disputed limitations in Claim 6, a claim that further limits the scope of Claim 1 by adding the requirement that the read/write memory units (main and spare) consist of “identical” disk drives. This limitation describes the preferred embodiment. (A 279.) (e) Claim 7 (Independent Claim): Disputed Limitations Claim 1 covers a computer system. Claim 7 relates to a method for performing the functions that are performed, inter alia, by the computer system described in Claim 1. Both parties agree that the limitations of independent claim 7 should be construed in the same manner as their counterpart limitations in independent claim 1, because the claims are analogues. Southwall Technologies, Inc. v. Cardinal IG Co., 54 F.3d 1570 (Fed.Cir.1995). I accept their invitation to construe the corresponding limitations in the two claims identically. (d) Claim 10 (Dependent Claim): Disputed Limitations As was true of claim 6, there are no disputed limitations in dependent claim 10. B. The ’979 Patent The specifications for the ’979 patent, like those for the ’342 patent, show a data storage system consisting of 39 disk drives, including 32 disks for storing data and seven disks for storing error correction code. The principal difference between the two patents is that the ’979 patent does not include the use of a spare memory unit to back up the main memory units. The only disputed claim is Claim 13. 1. Construction of Disputed Claims (a) Claim IS (Independent claim: Disputed Limitations) (i) A system for storing digital data words, each data word having a plurality of multi-bit data portions, and for storing said data words in response to a storage request received over a parallel data bus said system comprising ... The key disputed term in this part of the definition is “data word.” TM contends, without any particular intrinsic evidence to back its contention, that “data word” is no different than “data block” in the ’342 patent. The question that arises is why the inventor would use a different phrase to convey the same concept. The issue becomes more acute because the term “word” has a specific and particular meaning in computerese. It is a term of computer architecture, referring to the amount of data that can be transferred to or from input or output devices in one memory cycle. See S. Handel, A Dictionary of Electronics 392 (2d Ed.1966). The term “block of data,” as I have interpreted it in connection with the ’342 patent, contains no such specialized meaning and no such restriction. Therefore, I am unpersuaded that the two terms should be construed interchangeably. IBM would have me construe “data word” even more narrowly, as encompassing only groups of bits that are “small multiples of 32 bits,” on the ground that the largest “word” shown in the patent specifications is 256 bits, or 8 times 32 bits. IBM offers no persuasive reason why I should insert this limitation into the term “word,” although its counsel — while agreeing at oral argument that a “word” is indeed the amount of data that can be transferred in one memory cycle — suggested (without evidentiary support) that only a small amount of data could be transferred in a single memory cycle using current technology. This may or may not be true, but it is not necessary to interpolate so precise a limitation into the patent claim. Of course, I must interpret this limitation in light of the next following limitation, which specifically states that each data storage unit shall store “a respective one” of the several multi-bit data portions for each data word. (See below, page 32.) Thus, to fall within the literal patent claim, there must be an equivalent number of multi-bit data word portions and data storage units. That does not, however, mean that there may only be 32 data portions and corresponding 32 storage units in order to fall within the limitation. It is not inconceivable that a system could contain several hundred, or even thousand, data storage units, which would therefore be capable of storing a corresponding number of multi-bit word portions in parallel, one data portion to a memory unit. IBM would also have me declare that the bits of data within the word must be logically related in the sense that they are part of the same unit of information sent from the computer, and will be part of the same unit of information when returned to the computer. To illustrate its point: IBM contends that “Mary had a little lamb” could be a data word, because all the “bits” (that is to say, all the letters) are logically related; but “Mary had a little lamb and the Yankees are World Champions” could not, because this statement contains two ideas that are not logically related. TM argues that there is no such limitation in the patent claim, and points out that imposing any such limitation would exclude the preferred embodiment from the scope of the patent — a highly disfavored way of construing a patent, one that is “rarely, if ever, correct.” See Vi-tronics, 90 F.3d at 1583. TM is correct, and I decline to impose this restrictive definition on the claim. The parties agree that the term “parallel data bus” means a system containing more than one conductor within a computer along which information is transmitted from a source within the computer to a destination within that computer. And the term “multi-bit data portions” necessarily means data word subsets of two or more bits, because a system described in prior art could read only one bit at a time. (A 605.) I will thus describe this first limitation for the jury in the following terms: We are dealing in this ’979 patent with a system for storing something called digital data words. Digital means comprised or made up of digits, which as we already know can be either 0 or 1. And data, as we already know, means information- — -in this case, information that is encoded into digital form. Now exactly what is a data “word?” Well, the term “word,” in computerese, means the amount of data, or the number of bits (binary digits, or Os and Is), that can be transferred to or from input/output devices in a single memory cycle. It is not a specific number of bits; rather, it is whatever number of bits can be transferred to or from input/output devices in one memory cycle. We will hear a lot more about input and output devices and memory cycles as the trial wears on. In this particular system discussed in the ’979 patent, each digital data word is made up of two or more sub-parts, and each of those sub-parts contains more than one bit, that is, two or more bits. For example, if you had a data word that was 64 bits long, it might have 8 multi-bit data portions consisting of 8 bits each. Or it might have 32 multi-bit data portions consisting of 2 bits each. But however these sub-parts, these data portions, get whacked up, each one must have at least 2 bits. Now our system, to be comprehended within the ’979 patent, must have the capacity to store the data words in a particular way. The system has to be able to store the data words in response to a storage request that is received over something called a parallel data bus. When you think about a data bus, I want you to think about a Bee Line bus, or a city transit bus, or a school bus. Those are devices for transporting people from one place to another. Well, in a computer, a bus is a device for transporting data from one place to another. It consists of one or more conductors, along which information is transmitted from a source to a destination. And a parallel data bus is more or less exactly what it says: it is a data bus that has multiple conductors, or more than one wire, if you prefer, which can transmit data at the same time — in parallel, so to speak — from one place within a disk system to another. (ii) ... a plurality of data storage units each having a mechanically driven medium, said media of said different storage units being driven independently of each other, said data storage units storing data words with each data storage unit storing in parallel a respective one of said multi-bit data portions for each data word, each data storage unit including an arrangement for writing data portions onto its respective storage media ... This limitation deals with “data striping,” which is the salient storage feature of the ’979 patent. Data striping refers to a process of storing portions of a single unit of data (in this case, the previously defined “data word”) in multiple memory units. Striped data is literally spread out over several different storage media, so that all those media must be accessed in order to obtain the entire data unit. To offer a simple example: if the word “cat” were split into its three component parts, and each part were stored in a different memory unit, the word would be striped. This particular limitation calls for striping data so that each data storage unit in the system stores in parallel one multibit data portion of each data word. Thus, as noted above, the patent claim contemplates that there will be an equivalent number of data storage units and multi-bit data portions within a data word. The units in this limitation are specifically referred to as “data storage units.” IBM argues, as it did in connection with the ’342 patent, that, to fall within the patent, these storage units may store only data and may not contain error correction code. As was true with the ’342 patent, TM disagrees. I find that I must reach a different conclusion with the ’979 patent than I did with the ’342 patent on this particular point, because there are two significant differences between the two sets of claims. First and foremost, the plain language of the ’979 claim is not the same as the language of the ’342 claim. The ’979 language identifies two types of storage units: “data storage units” (this limitation) and “correction bit storage units” (the next limitation). Second, confirming that the inventor intended the plain meaning of the words he used in drafting the claim, there are no fewer than five prior art distinctions in the patent on the ground that the disks in the ’979 invention are dedicated either to data or to memory, while the disks in prior art were not so dedicated. To take one example, in a Preliminary Amendment following the rejection of its original claims, the patentee distinguished a prior art reference in an article by Professors Haran Boral and David J. DeWitt, entitled Database Machines: An Idea Whose Time Has Passed? (A 725), by noting that the article did “not suggest providing a separate error correction storage device ... (A 670.) In another submission, TM advised the Patent Examiner that “One general feature of applicants’ invention (recited in claim 1) is in storing a data word on one mechanically driven medium (i.e., a disk drive) and storing one or more error correction bits for the data word independently on a medium that is mechanically driven independently from the medium on which the data word is stored (e.g., on a separate disk drive) ... ” (emphasis added). (A 440.) Finally, the parties disagree about the meaning of the term “storing in parallel” as used in this particular limitation. IBM contends that a necessary construction of the patent is that none of the data storage units can operate independently of the others; they must all be accessed simultaneously (synchronously), in a single operation, in order to store the data word. In support of its argument, IBM points out that, at column five, lines 24-28 of the patent, the inventor writes, “The storage and retrieval of data via all of the SCSI buses to the controllers occurs synchronously in parallel. That is, from the point of view of the bus adaptor, related data is passed in a single operation over all of the SCSI buses at the same time.” TM argues that the phrase “storing in parallel” should be read literally, as applying only to storage, which TM concedes must be across multiple storage media. It urges that this phrase neither literally says nor fairly implies anything about how data are retrieved and contends that data may be accessed asynchronously as well as synchronously within the literal terms of the patent. TM also points out that Thinking Machines’s response to the Patent Examiner distinguished a prior art reference to the Boral article on the basis that the article (unlike TM’s application) did not permit “data transfer which may be asynchronous.” (A 608.) The phrase “storing in parallel” is discussed extensively in the file history, primarily by way of differentiating- the ’979 patent from other inventions of the time. It appears that the Examiner allowed the claims as proposed because of the addition of the words “in parallel” to the claim in question. Construction of this term, then, would appear to be key to this invention. I find it to be the most difficult question of all those presented to me, precisely because the file wrapper contains references that can be read to support either point of view. The Court’s expert, Professor Lipovski, has written a letter that affords valuable assistance in this regard. Professor Li-povski notes that the SCSI protocol— which is described in the patent (and therefore constitutes intrinsic evidence), and which all parties agree is pertinent to the reading and writing of data under this patent — does not have a command to synchronize multiple disks so as to read or write data in parallel (i.e., at exactly the same time). It does, however, have commands that can be sent to multiple disks so that data can be sent to or from the disks on which they are stored using a common clock (i.e., synchronously). This suggests that IBM’s construction — which relies upon the phrase “from the point of view of the bus adapter, related data (i.e., a data word) is passed in a single operation over all of the SCSI buses at the same time” (the SCSI buses are the connections between the adapter and the disk drive units) — is correct. Moreover, because the SCSI protocol as contained in the patent lacks a command to synchronize multiple disks for reading and writing data in parallel, IBM’s construction is not inconsistent with the prior art reference to the Boral article, which distinguished the ’979 application as permitting asynchronous data transfer. Thus, this second limitation will be described to the jury in the following terms: Second, in the ’979 patent, the data portions have to be “striped” in order to be stored. Now just what is “data striping?” Well, when you stripe data, you split it up so that sub-portions of the basic unit of data are stored in different storage units. To take a very simple example, if we think of the word “cat” as a data word, we would stripe that data word by taking it apart and placing one letter on each of three different disk drives. In order to figure out what the data word said, we would then have to read all three disk drives. By contrast, if you did not stripe the data, you would store all three letters of the word “cat” on the same disk drive. So in this ’979 patent, we have two or more data storage units, each one mechanically driven independently of the others, and each one of those data storage units stores one of those multi-bit portions of our data word. The multi-bit portions of the data word are thus said to be stored “in parallel.” For this purpose, the data storage units have to have some way to write the data portions onto their storage media. The patent claim specifies that each data storage unit stores “a respective one” multi-bit word portion. “A respective one” means exactly one, no more and no less. That means there must be an equal number of data storage units and multi-bit word portions in a system covered by the ’979 patent. And the system needs to have the ability to transfer all of these multi-bit word portions over the connections between the disk drives and the adapter — the so-called SCSI buses — at the same time, (iii) ... at least one correction bit storage unit including a storage medium that is mechanically driven independently from the mechanically-driven media on which said data word is stored, for storing an error correction code associated with said data words, said correction bit storage unit including as arrangement for writing correction bits onto its storage media The construction of this limitation necessarily follows from certain conclusions reached earlier. Because the claim clearly distinguishes between data storage units and error correction code units, I must advise the jury that the “correction bit storage unit” stores corrections bits, not any underlying data (even data that are not associated with a particular correction bit). The other disputed element of this limitation concerns the meaning of the phrase “error correction code.” IBM argues that this phrase has the same meaning as in the ’842 patent, namely, it must both detect and correct data. TM says no, that this patent claim, unlike the ’342 claim, says nothing about detecting as well as correcting data. This issue assumes particular importance