{
  "patent_number": "US 10956815",
  "country": "US",
  "title": "How to Fix Faulty Memory Cells in AI Chips",
  "original_title": "Killing asymmetric resistive processing units for neural network training",
  "summary": "This patent describes a system that tests individual memory cells in AI chips for uneven behavior and then permanently disables the faulty ones before the chip starts learning, making AI training more efficient.",
  "what_it_does": "The patent describes a system for improving neural network training on specialized hardware called Resistive Processing Unit (RPU) arrays. An RPU array has many tiny memory cells (RPUs) arranged in a grid, where each RPU's electrical state (its \"conduction state\") stores a \"weight\" for the AI. Before training begins, a controller measures each RPU's \"asymmetry value\" by sending a positive electrical pulse and a negative electrical pulse and comparing how much the RPU's conduction state changes for each (Claim 1). If an RPU's asymmetry is too high, meaning it behaves differently depending on the electrical direction, the system \"burns\" it by applying a high voltage to permanently disable it (Claim 2, Claim 6). This ensures only reliable RPUs are used for training.",
  "what_it_does_not_cover": [
    "Does not cover systems that train neural networks on traditional silicon chips like CPUs or GPUs, as it specifically targets RPU arrays.",
    "Does not cover methods of improving RPU array performance that don't involve measuring and disabling asymmetric RPUs.",
    "Does not cover RPU arrays where faulty units are simply ignored or remapped instead of being physically \"burned\" by a high voltage.",
    "Does not cover identifying faulty RPUs *during* or *after* the neural network training process.",
    "Does not cover RPUs that store information without also locally performing data processing operations (Claim 9)."
  ],
  "filed": "2017-05-31",
  "granted": "2021-03-23",
  "expires": "2037-05-31",
  "status": "active",
  "holder": "International Business Machines",
  "holder_url": "https://patentbrief.org/company/international-business-machines",
  "inventors": [
    {
      "name": "Tayfun Gokmen",
      "url": "https://patentbrief.org/inventor/tayfun-gokmen"
    }
  ],
  "times_cited": 3,
  "tags": [
    "semiconductors",
    "ai_ml",
    "consumer_electronics",
    "telecommunications"
  ],
  "abstract": "Technical solutions are described for improving efficiency of training a resistive processing unit (RPU) array using a neural network training methodology. An example method includes reducing asymmetric RPUs from the RPU array by determining an asymmetric value of an RPU from the RPU array, and burning the RPU in response to the asymmetry value being above a predetermined threshold. The RPU can be burned by causing an electric voltage across the RPU to be above a predetermined limit. The method further includes initiating the training methodology for the RPU array after the asymmetric RPUs from the RPU array are reduced.",
  "url": "https://patentbrief.org/patent/us/10956815/killing-asymmetric-resistive-processing-units-for-neural-network-training",
  "markdown_url": "https://patentbrief.org/patent/us/10956815/killing-asymmetric-resistive-processing-units-for-neural-network-training/md",
  "google_patents_url": "https://patents.google.com/patent/US10956815",
  "relatedPatents": [
    {
      "patentNumber": "10248907",
      "countryCode": "US",
      "title": "How a Single Electronic Component Can Learn and Process AI Data",
      "url": "https://patentbrief.org/patent/us/10248907/resistive-processing-unit"
    },
    {
      "patentNumber": "11741188",
      "countryCode": "US",
      "title": "How a Chip Uses Memory to Speed Up AI Calculations",
      "url": "https://patentbrief.org/patent/us/11741188/hardware-accelerated-discretized-neural-network"
    },
    {
      "patentNumber": "10410117",
      "countryCode": "US",
      "title": "How to Save and Reuse Skills Learned by Artificial Intelligence Hardware",
      "url": "https://patentbrief.org/patent/us/10410117/method-and-a-system-for-creating-dynamic-neural-function-libraries"
    },
    {
      "patentNumber": "10832138",
      "countryCode": "US",
      "title": "How to Automatically Expand Neural Networks by Adding New Nodes",
      "url": "https://patentbrief.org/patent/us/10832138/gpt-language-model-pre-training"
    },
    {
      "patentNumber": "12423586",
      "countryCode": "US",
      "title": "Making AI Smarter by Focusing on Unsure 'Nodes'",
      "url": "https://patentbrief.org/patent/us/12423586/training-nodes-of-a-neural-network-to-be-decisive"
    }
  ]
}