公开(公告)号：US20250117915A1

公开(公告)日：2025-04-10

申请号：US18482209

申请日：2023-10-06

Applicant: Applied Materials, Inc.

Inventor： Navneet Kumar Singh ,  Arun Ramaswamy Srivatsa ,  Sachin Dangayach ,  Zvi Hersh Goldshtein ,  Rahul Reddy Komatireddi ,  Sutapa Dutta ,  Arv Nagpal ,  Yen-Tien Wu

IPC: G06T7/00

Abstract: Implementations disclosed describe, among other things, a systems and techniques for perform efficient inspection of a semiconductor manufacturing sample. The techniques include collecting optical inspection data for training sample(s) that have a plurality of defects. The techniques further include generating, using the optical inspection data, a training data set that includes descriptions, images, and ground truth classifications for the defects. The techniques further include using the training data set to train a plurality of machine learning (ML) classifiers to generate predicted classifications for the defects in the training sample(s). The techniques further include selecting, using the predicted classifications and the ground truth classifications, one or more ML classifiers that meet one or more accuracy criteria, and using the selected ML classifier(s) to classify defects in the semiconductor manufacturing sample.

公开(公告)号：US20240403258A1

公开(公告)日：2024-12-05

申请号：US18328852

申请日：2023-06-05

Applicant: Applied Materials, Inc.

Inventor： Bilal Shafi Sheikh ,  Tameesh Suri ,  Nathaniel See ,  Sutapa Dutta ,  Yun-Ting Sun ,  Udaykumar Diliprao Hanmante ,  Naveed Zaman

IPC: G06F15/78 ,  G06N5/04

Abstract: A chiplet-based architecture may quantize, or reduce, the number of bits at various stages of the data path in an artificial-intelligence processor. This architecture may leverage the synergy between quantizing multiple dimensions together to greatly decrease the memory usage and data path bandwidth. Internal weights may be quantized statically after a training procedure. Accumulator bits and activation bits may be quantized dynamically during an inference operation. New hardware logic may be configured to quantize the outputs of each operation directly from the core or other processing node before the tensor is stored in memory. Quantization may use a statistic from a previous tensor for a current output tensor, while also calculating a statistic to be used on a subsequent output tensor. In addition to quantizing based on a statistic, bits can be further quantized using a Kth percentile clamping operation.