DCT
3:18-cv-04571
Carnegie Mellon University v. LSI Corp
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Carnegie Mellon University (Pennsylvania)
- Defendant: LSI Corporation (Delaware) and Avago Technologies U.S. Inc. (Delaware)
- Plaintiff’s Counsel: K&L Gates LLP
- Case Identification: 3:18-cv-04571, N.D. Cal., 07/27/2018
- Venue Allegations: Venue is alleged to be proper as Defendants are headquartered in the district, maintain a regular and established place of business, and have committed acts of infringement within the district.
- Core Dispute: Plaintiff alleges that Defendants’ read channel chips for hard disk drives infringe patents related to methods for improving data-detection accuracy by using signal-dependent and correlation-sensitive techniques.
- Technical Context: The technology addresses the challenge of accurately reading data from high-density magnetic storage media, where signal noise is a significant limiting factor.
- Key Procedural History: The complaint notes that the asserted claims of the patents-in-suit were previously litigated against Marvell Technology Group, resulting in a jury verdict of infringement and validity, which was affirmed on appeal, and ultimately settled for $750 million. The complaint also states that the asserted claims were confirmed as patentable by the USPTO in ex parte reexamination proceedings requested in 2014.
Case Timeline
| Date | Event |
|---|---|
| 1997-05-09 | Priority Date for ’839 and ’180 Patents |
| 2001-03-13 | U.S. Patent No. 6,201,839 Issues |
| 2002-08-20 | U.S. Patent No. 6,438,180 Issues |
| 2003-08-05 | CMU allegedly notifies LSI’s predecessor (Agere) of the Patents |
| 2003-09-01 | Agere allegedly introduces "Data-dependent NPML" technology |
| 2009-12-31 | LSI executive allegedly admits to using inventor’s "detector" (approx. date) |
| 2010-09-20 | LSI executive allegedly admits in email LSI uses inventor’s algorithms |
| 2014-01-21 | Ex parte reexamination requested for the Patents-in-Suit |
| 2014-10-29 | USPTO confirms patentability of ’839 Patent claim 4 |
| 2015-01-30 | USPTO confirms patentability of ’180 Patent claims 1 and 2 |
| 2017-09-01 | CMU sends notice letter to Defendants regarding infringement |
| 2018-07-27 | Complaint Filed |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 6,201,839 - "Method and Apparatus for Correlation-Sensitive Adaptive Sequence Detection," Issued March 13, 2001
The Invention Explained
- Problem Addressed: The patent identifies a problem in high-density magnetic recording where noise from various sources (media, equalizers, nonlinearities) is not simple, but rather is "heavily correlated" between adjacent signal samples and its characteristics are "signal dependent" (i.e., the noise pattern depends on the data pattern being read) ('839 Patent, col. 1:38-41, col. 2:5-14). Conventional detectors that assume simple, uncorrelated noise suffer significant performance degradation at high densities ('839 Patent, col. 1:60-64).
- The Patented Solution: The invention proposes a detector that computes branch metrics for a Viterbi-like algorithm in a "correlation-sensitive" manner. It uses branch metric functions that account for both the signal-dependent nature of the noise and the correlation between noise samples ('839 Patent, Abstract). To handle variations from track to track, the system adaptively tracks the noise statistics (e.g., covariance matrices) using a feedback loop that incorporates past detector decisions and signal samples ('839 Patent, col. 2:15-24, Fig. 2).
- Technical Importance: By creating a model that more accurately reflects the complex, correlated, and signal-dependent nature of noise in real-world recording channels, the invention enables more accurate data detection, which in turn allows for higher storage densities ('839 Patent, col. 2:25-29).
Key Claims at a Glance
- The complaint asserts independent claim 4 ('839 Patent, col. 14:10-19; Compl. ¶174).
- Claim 4 Elements: A method of determining branch metric values for a Viterbi-like detector, comprising:
- selecting a branch metric function for each of the branches at a certain time index from a set of signal-dependent branch metric functions;
- applying each of said selected functions to a plurality of signal samples to determine the metric value corresponding to the branch for which the applied branch metric function was selected;
- wherein each sample corresponds to a different sampling time instant.
U.S. Patent No. 6,438,180 - "Soft and Hard Sequence Detection in ISI Memory Channels," Issued August 20, 2002
The Invention Explained
- Problem Addressed: Similar to its parent, the ’180 Patent addresses detecting data in channels with intersymbol interference (ISI) where the noise is complex—specifically, it can be correlated, signal-dependent, or both ('180 Patent, col. 2:19-24). The patent notes that existing soft-output detectors, which are important for advanced error-correction schemes, are typically designed only for simple "white noise" channels ('180 Patent, col. 2:19-24).
- The Patented Solution: The invention describes a method for determining branch metric values that is suitable for channels with such complex noise. The method involves receiving time-variant signal samples with these noise characteristics, selecting an appropriate branch metric function, and applying it to the samples ('180 Patent, Abstract). This approach is general enough to be used in both hard-output detectors (like Viterbi) and soft-output detectors (like BCJR), enabling the use of advanced iterative decoding ("turbo") techniques in realistic, non-ideal channels ('180 Patent, col. 2:1-18, col. 13:10-14, Fig. 14).
- Technical Importance: The invention provides a framework for designing more robust soft-output detectors, which generate reliability information instead of just hard decisions, allowing powerful error-correction codes to function effectively in real-world, high-density storage channels with complex noise characteristics ('180 Patent, col. 13:5-14).
Key Claims at a Glance
- The complaint asserts independent claim 1 and dependent claim 2 ('180 Patent, col. 15:39-51; Compl. ¶185).
- Claim 1 Elements: A method of determining branch metric values in a detector, comprising:
- receiving a plurality of time variant signal samples, the signal samples having one of signal-dependent noise, correlated noise, and both signal dependent and correlated noise associated therewith;
- selecting a branch metric function at a certain time index; and
- applying the selected function to the signal samples to determine the metric values.
- Claim 2 further specifies that "the branch metric function is selected from a set of signal-dependent branch metric functions."
III. The Accused Instrumentality
Product Identification
- The complaint identifies "HDD read channel devices," including "HDD Chips" sold under the "TrueStore" tradename and associated "Simulators" (collectively, "Exemplary Accused Products") (Compl. ¶79). Specific product families named are the RC5101 Spyder, RC5110 Sypder, and RC5200 Spyder read channels and SoCs (Compl. ¶79).
Functionality and Market Context
- The accused products are integrated circuits within hard disk drives (HDDs) that process the noisy analog "readback" signal from the magnetic disk to recover the original digital data (Compl. ¶¶33, 37). The complaint alleges these products implement the patented methods by using what the industry refers to as "pattern-dependent filters" or "data dependent noise prediction" (DDNP) (Compl. ¶¶89-92). This functionality is alleged to be implemented via a bank of Finite Impulse Response (FIR) filters, where each filter is tuned to a specific data pattern, thereby accounting for signal-dependent noise (Compl. ¶¶104, 107). The complaint alleges that this technology is crucial for achieving competitive bit-error-rates (BER) and that Defendants adopted it to compete with rivals, like Marvell, who had already adopted the technology (Compl. ¶¶100, 130-131). The complaint also describes a "design win" sales cycle, where Defendants work closely with HDD manufacturers to have their chips designed into future products (Compl. ¶¶138-139).
IV. Analysis of Infringement Allegations
The complaint alleges that Defendants' TrueStore products perform the patented methods by implementing a function known in the industry as "data dependent noise prediction" (DDNP) or a "pattern-dependent filter" (Compl. ¶¶89, 92). A 2012 presentation slide, allegedly from an LSI executive, identifies "pattern-dependent filter" as a "major" technology for "noise processing" (Compl. ¶89). A separate slide from the same presentation shows a read channel block diagram containing a block labeled "Equalizer & DDNP," which the complaint asserts refers to the patented technology (Compl. ¶¶91-92). Further, the complaint points to a 2004 roadmap from Agere (LSI's predecessor) showing the introduction of "Data-dependent NPML" in its "Redtail" read channel, which is alleged to be an industry term for the patented methods (Compl. ¶¶97-98). Finally, the complaint references a patent application assigned to LSI that shows a "DDNP Filter Bank" where each filter is "tuned to a respective, defined data pattern" to estimate noise correlation (Compl. ¶104).
'839 Patent Infringement Allegations
| Claim Element (from Independent Claim 4) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| selecting a branch metric function for each of the branches at a certain time index from a set of signal-dependent branch metric functions | The accused products' "DDNP" feature allegedly uses a bank of data-dependent noise predictive filters, where each filter is tuned to a specific data pattern. This is alleged to be the selection of a function from a set of signal-dependent functions. | ¶¶90, 107, 114 | col. 2:15-20 |
| and applying each of said selected functions to a plurality of signal samples...wherein each sample corresponds to a different sampling time instant | The selected FIR filter is applied to multiple, time-variant signal samples from the readback signal to determine the branch metric value. The FIR filters are implemented with multiple taps, each for a separate time-variant sample. | ¶¶58, 107, 114 | col. 7:51-55 |
'180 Patent Infringement Allegations
| Claim Element (from Independent Claim 1, as limited by Claim 2) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| receiving a plurality of time variant signal samples, the signal samples having one of signal-dependent noise, correlated noise, and both...associated therewith | The accused products operate on the readback signal from a high-density HDD, which the complaint alleges inherently contains signal-dependent and correlated "media noise." | ¶¶48-50, 114 | col. 2:19-24 |
| selecting a branch metric function...from a set of signal-dependent branch metric functions | As with the '839 patent, the "DDNP" or "pattern-dependent filter" functionality allegedly selects a specific noise-predictive filter from a bank, which corresponds to selecting a function from a set of signal-dependent branch metric functions. | ¶¶90, 107, 114 | col. 2:32-34 |
| and applying the selected function to the signal samples to determine the metric values | The selected FIR filter is applied to the received signal samples to compute the branch metric values for the detector. | ¶¶58, 107, 114 | col. 15:45-48 |
- Identified Points of Contention:
- Technical Questions: A primary question will be whether the functionality labeled "DDNP" or "pattern-dependent filter" in Defendants' internal documents and products actually performs the steps recited in the claims. The analysis will require evidence mapping the operation of the accused FIR filter banks to the claimed steps of "selecting" and "applying" a "signal-dependent branch metric function."
- Scope Questions: The dispute may center on the definition of "selecting a branch metric function from a set." It raises the question of whether using a bank of filters, each tuned to a data pattern, constitutes "selecting" from a "set" of functions, or if it constitutes a single, larger, adaptable function whose parameters change.
V. Key Claim Terms for Construction
The Term: "signal-dependent branch metric functions" (from '839 claim 4 and '180 claim 2)
- Context and Importance: This term is the technological core of the asserted claims. The outcome of the infringement analysis depends heavily on whether the accused "DDNP" systems are found to use functions that are "signal-dependent" as the patent defines it.
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The specification describes the concept broadly, stating that in magnetic recording, "the functional form of joint conditional pdf...is different for different symbol sequences" ('839 Patent, col. 3:24-29). This could support an interpretation where any metric calculation that systematically changes based on the data pattern qualifies.
- Evidence for a Narrower Interpretation: The specification provides a specific mathematical expression for the "general correlation-sensitive metric," which involves covariance matrices and their sub-matrices ('839 Patent, col. 6:45-52, eq. 11). This detailed embodiment might be used to argue that the term should be limited to functions having this particular mathematical structure.
The Term: "selecting a branch metric function" (from '839 claim 4 and '180 claim 1)
- Context and Importance: This term recites an active step of choosing a function. Practitioners may focus on this term because the infringement theory relies on the idea that choosing a specific filter from a "DDNP filter bank" is equivalent to "selecting" a function.
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The patent's description of different covariance matrices for "each branch of the tree/trellis" could support the view that any system that applies different processing logic (e.g., different FIR filter coefficients) for different data patterns is performing a "selection" ('839 Patent, col. 2:18-20).
- Evidence for a Narrower Interpretation: The language could be argued to require a discrete choice from a pre-defined "set" of distinct functions, rather than the continuous adaptation of a single function's parameters or the routing of a signal through one of several parallel hardware paths.
VI. Other Allegations
- Indirect Infringement: The complaint alleges both active inducement and contributory infringement. The inducement allegation is based on Defendants allegedly providing customers with product specifications, user guides, register maps, and technical briefings that instruct them on how to enable and use the accused DDNP features (Compl. ¶¶117, 123, 177). The contributory infringement allegation is based on the assertion that the accused read channel components are especially made for practicing the invention, are a material part of it, and have no substantial non-infringing use (Compl. ¶¶128, 178).
- Willful Infringement: The complaint alleges that Defendants had knowledge of the patents and their infringement from multiple sources, including: a 2003 notice letter sent to their predecessor, Agere (Compl. ¶153); alleged direct admissions of use by a senior LSI executive to one of the inventors in 2009 and 2010 (Compl. ¶¶157-161); citations to the patents-in-suit during the prosecution of Defendants' own patents (Compl. ¶¶163-166); and knowledge of the widely-publicized CMU v. Marvell litigation involving the same patents (Compl. ¶¶168-170). A final notice letter was also allegedly sent in September 2017 (Compl. ¶172).
VII. Analyst’s Conclusion: Key Questions for the Case
This case presents a dispute grounded in sophisticated signal processing technology, with a significant history of prior litigation and alleged notice. The outcome will likely depend on the court's resolution of several key questions:
- A central issue will be one of evidentiary linkage: Can the industry terminology and high-level block diagrams cited in the complaint, such as "DDNP" and "pattern-dependent filter," be proven to map directly onto the specific technical and mathematical requirements of the asserted claims? The case may turn on whether CMU can show that the accused products' actual operations match the patented methods, not just the labels used to describe them.
- A second core issue will be one of definitional scope: How will the court construe the term "selecting a branch metric function from a set"? The viability of the infringement case rests on whether this phrase can be interpreted to cover the alleged operation of the accused products—activating a particular FIR filter from a bank based on a data pattern—or if it requires a more discrete choice between formally distinct mathematical functions.
- Given the extensive allegations of pre-suit knowledge, a key question for willfulness and enhanced damages will be one of scienter: Can Defendants successfully argue that they had a good-faith belief of non-infringement or invalidity, despite the prior Marvell litigation, the reexamination outcomes, and the alleged admissions by their own executive?