Microsoft Corp v. ThroughPuter Inc

1. Case Identification

• Case #: IPR2023-00852
• Patent #: 11,150,948
• Filed: April 17, 2023
• Petitioner(s): Microsoft Corporation
• Patent Owner(s): ThroughPuter, Inc.
• Challenged Claims: 1-15

2. Patent Overview

• Title: Managing Programmable Logic-Based Processing Unit Allocation on a Parallel Data Processing Platform
• Brief Description: The ’948 patent describes techniques for managing and allocating programmable logic-based processing units (like FPGAs) on a parallel, multi-stage, manycore processor platform. The system dynamically adds processing units to program instances in response to increased demand, which is monitored based on characteristics like load or utilization.

3. Grounds for Unpatentability

Ground 1: Obviousness over Nollet, Tuan, and Seo - Claims 1-15 are obvious over Nollet in view of Tuan and Seo.

• Prior Art Relied Upon: Nollet (Application # 2009/0187756), Tuan (a 2009 study on reconfigurable processors), and Seo (a 2008 IEEE article on energy-efficient task scheduling).

• Core Argument for this Ground:

  • Prior Art Mapping: Petitioner argued that the combination of Nollet, Tuan, and Seo teaches every limitation of the challenged claims. Nollet discloses the foundational multi-processor computing platform with a plurality of programmable logic units (FPGAs) capable of running multiple different programs (e.g., a movie player and a 3D game) concurrently. Nollet further teaches reallocating resources based on parameters like the load of a computing unit. However, Nollet leaves the specific implementation details of task mapping and resource allocation open.

  Petitioner asserted Tuan fills this gap by teaching how to map a complex application onto a reconfigurable, tile-based architecture as a "multi-stage program instance" with inter-task communication paths (using FIFOs), which improves throughput. The combination of Nollet and Tuan thus allegedly discloses a parallel processing platform that accelerates services by forming multiple multi-stage program instances connected by inter-task communication paths using reconfigurable cross-connects, as recited in independent claims 1, 6, and 11.

  Seo was introduced to provide the specific "adaptive optimized resource allocation policy" claimed. Seo describes a dynamic core scaling algorithm for multicore processors that optimizes power consumption while meeting performance demands. This algorithm monitors characteristics of the system—specifically, task utilization and execution time (i.e., load)—to determine the optimal number of active cores. When demand increases, Seo’s algorithm adds more processing cores. Petitioner contended this directly teaches the limitations in claims 1, 6, and 11 of monitoring a characteristic like load and, in response to increased demand, adding one or more processing units to the program instance. The policy is also shown to relate to balancing load, as required by dependent claims 2, 7, and 12.

  • Motivation to Combine: A Person of Ordinary Skill in the Art (POSITA) would combine Nollet’s general framework for a reconfigurable computing platform with Tuan's specific methods for mapping applications into multi-stage tasks to improve throughput and manage inter-task communication. Because Nollet and Tuan both identify the need to balance performance with power consumption but do not provide a specific algorithm, a POSITA would have been motivated to look to known solutions like Seo. Seo’s dynamic core scaling algorithm, which explicitly manages the trade-off between performance and power by adding or removing cores based on monitored load, provides a direct and logical implementation for the systems of Nollet and Tuan.
  • Expectation of Success: Petitioner argued a POSITA would have a reasonable expectation of success because the combination involves applying known techniques (Tuan's task mapping and Seo's dynamic scaling) to a known type of system (Nollet's reconfigurable platform) to achieve predictable results. Each component would perform its known function to improve performance and energy efficiency.

4. Arguments Regarding Discretionary Denial

• Petitioner argued that discretionary denial under Fintiv is inappropriate because the related district court litigation has been stayed pending the resolution of Microsoft's IPR petitions. This stay, granted early in the litigation before significant investment by the court or parties, allays any concerns about inefficiency or duplication of efforts, weighing heavily in favor of institution.
• Petitioner further argued that denial under 35 U.S.C. §325(d) is unwarranted because the core prior art references, Tuan and Seo, are new and were never considered by the USPTO during prosecution of the ’948 patent. The petition asserted that these references are materially different from the single, non-relied-upon reference of record from prosecution and that they teach the critical limitations of dynamic resource allocation based on monitored load, which was a key aspect of the invention.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 1-15 of the ’948 patent as unpatentable.