MediaTek had 20 patents in artificial intelligence during Q4 2023. MediaTek Inc filed patents related to optimizing neural network models for fixed-point arithmetic execution, voice wakeup detection devices with adjustable gain and weight values, a method for 3D pooling operations on input and output tensors, using neural networks for flexible block placement in IC design, and using neural networks for macro placement in IC design with action masks for optimizing placement based on design rules. GlobalData’s report on MediaTek gives a 360-degreee view of the company including its patenting strategy. Buy the report here.

MediaTek grant share with artificial intelligence as a theme is 5% in Q4 2023. Grant share is based on the ratio of number of grants to total number of patents.

Recent Patents

Application: Compiling asymmetrically-quantized neural network models for deep learning acceleration (Patent ID: US20230401420A1)

The patent filed by MediaTek Inc. describes a system and method for neural network computing that involves receiving a neural network model with asymmetric operations, compiling these operations into symmetric operations with combined bias values, and generating a symmetric neural network model for execution in fixed-point arithmetic. The method includes converting zero-padding in convolution layers to zero point padding, mapping weights in convolution layers from floating-point to fixed-point, and merging scaling factors of inputs, outputs, and weights in asymmetric convolutions. The system consists of a memory storing a compiler and neural network models, and processing hardware to execute the compiled symmetric operations on an accelerator circuit.

The patent also details specific claims related to different types of asymmetric operations within neural networks, such as convolution, parametric ReLU (PReLU), and addition. It outlines the computation of combined bias values for these operations based on zero points of inputs and outputs, as well as multiplicative factors. The system and method can be implemented using offline or online compilers, depending on whether the neural network model is compiled before or after the execution of the computer program. Overall, the patent focuses on optimizing neural network computations by converting asymmetric operations to symmetric operations with combined bias values for efficient execution in fixed-point arithmetic.

To know more about GlobalData’s detailed insights on MediaTek, buy the report here.

Premium Insights

From

The gold standard of business intelligence.

Blending expert knowledge with cutting-edge technology, GlobalData’s unrivalled proprietary data will enable you to decode what’s happening in your market. You can make better informed decisions and gain a future-proof advantage over your competitors.

GlobalData

GlobalData, the leading provider of industry intelligence, provided the underlying data, research, and analysis used to produce this article.

GlobalData Patent Analytics tracks bibliographic data, legal events data, point in time patent ownerships, and backward and forward citations from global patenting offices. Textual analysis and official patent classifications are used to group patents into key thematic areas and link them to specific companies across the world’s largest industries.