PDF Solutions Inc v. Ocean Semiconductor LLC

1. Case Identification

• Case #: IPR2022-00671
• Patent #: 8,676,538
• Filed: March 9, 2022
• Petitioner(s): PDF Solutions, Inc.
• Patent Owner(s): Ocean Semiconductor LLC
• Challenged Claims: 1-16, 18-26, 28-31

2. Patent Overview

• Title: System and Method for Dynamic Fault Detection in Semiconductor Manufacturing
• Brief Description: The ’538 patent discloses a method for improving fault detection in workpiece processing, such as semiconductor manufacturing. The system performs a fault detection analysis, determines a relationship between a process parameter and a detected fault, and dynamically adjusts a weighting of that parameter based on the determined relationship for use in subsequent fault detection analyses.

3. Grounds for Unpatentability

Ground 1: Claims 1, 2, 6, 9-13, 18-19, 21-23, 28, and 29 are anticipated by Brcka under 35 U.S.C. §102.

• Prior Art Relied Upon: Brcka (Application # 2006/0259198).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Brcka discloses every limitation of the challenged claims. Brcka teaches an "intelligent modeling method" using a neural network to monitor, analyze, and predict failures in semiconductor processing equipment. Petitioner asserted this system meets independent claims 1 and 18 by: (a) performing a fault detection analysis on a workpiece by collecting processing measurements; (b) determining a relationship between input parameters (process measurements) and a detected fault (maintenance or failure events) by training the neural network model; (c) adjusting the "weight values" associated with the network's connections based on this relationship through a continuous "learning" process triggered by a calculated "reward value" when a fault occurs; and (d) applying the adjusted weights to analyze subsequent workpieces in a continuous feedback loop. Petitioner further argued Brcka anticipates the dependent claims by explicitly teaching the processing of semiconductor wafers, using parameters such as process pressure and chuck temperature, employing a neural network as a fault detection model, using metrology and tool state data as inputs, and performing principal component analysis (PCA) to simplify data for the model.

Ground 2: Claims 3-5, 7, 8, 14-16, 20, 24-26, 30, and 31 are obvious over Brcka in view of POSA Knowledge under 35 U.S.C. §103.

• Prior Art Relied Upon: Brcka (Application # 2006/0259198) in view of the general knowledge of a Person of Ordinary Skill in the Art (POSA).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner contended that the remaining claims would have been obvious extensions of Brcka's teachings to a POSA knowledgeable in neural networks and process control. A POSA would have understood that the "weight values" in Brcka's neural network inherently represent the "causation" or "importance" of an input parameter on a detected fault, thereby rendering claims 3, 4, 30, and 31 obvious. It would have been an obvious modification to designate certain faults as "significant" (claim 5) to prioritize more costly events, a common goal in process control. Similarly, a POSA would have understood that the weight value signifies whether a parameter is a "significant factor" (claims 7, 20). Adjusting weights by "increasing" or "decreasing" them (claims 8, 14, 24, 26) is the fundamental and obvious mechanism for implementing Brcka's "learning" rule. Finally, a POSA would know that increasing a parameter's weight could be used to adjust the model's sensitivity, requiring a smaller fluctuation to trigger a fault (claims 15, 25), and that for parameters with asymmetric acceptable ranges, this could also require a larger fluctuation in another direction (claim 16).
  • Motivation to Combine: A POSA would be motivated to apply their general knowledge of neural network functionality to refine Brcka's disclosed system. The motivation was to enhance the system's practicality and robustness by implementing well-known concepts like fault prioritization, parameter significance, and sensitivity tuning, which were common and desirable goals in the field of advanced semiconductor fault detection.
  • Expectation of Success: A POSA would have had a high expectation of success in making these modifications. The proposed changes involved applying standard, predictable principles of neural network operation (e.g., interpreting weight values, adjusting learning rules) and process control theory to the specific framework disclosed by Brcka, rather than requiring any undue experimentation or invention of new techniques.

4. Arguments Regarding Discretionary Denial

• Petitioner argued against discretionary denial under both 35 U.S.C. §314(a) (Fintiv) and 35 U.S.C. §325(d). Regarding Fintiv, Petitioner asserted that it is not a party to any parallel district court litigation involving the ’538 patent, that trial dates in those cases are uncertain and have historically been subject to delay, and that there is no overlap in the invalidity arguments as invalidity contentions had not been served. Regarding §325(d), Petitioner argued that the primary prior art reference, Brcka, was never presented to or considered by the USPTO during the original prosecution. Therefore, the petition raises new art and arguments that warrant the Board's consideration.

5. Relief Requested

• Petitioner requests institution of an inter partes review (IPR) and cancellation of claims 1-16, 18-26, and 28-31 of the ’538 patent as unpatentable.