17983058. NEUROMORPHIC DEVICE simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)

NEUROMORPHIC DEVICE

Organization Name

SAMSUNG ELECTRONICS CO., LTD.

Inventor(s)

Youngnam Hwang of Suwon-si (KR)

NEUROMORPHIC DEVICE - A simplified explanation of the abstract

This abstract first appeared for US patent application 17983058 titled 'NEUROMORPHIC DEVICE

Simplified Explanation

The abstract describes a neuromorphic device that consists of multiple cell tiles, each containing a cell array with memory cells storing weights of a neural network. The device also includes row drivers, cell analog-digital converters (ADCs), and a controller. The controller selects valid cell tiles, performs neural network-based arithmetic operations, and redundantly stores weights of a specific layer in multiple valid cell tiles.

• The neuromorphic device is designed to mimic the structure and functionality of the human brain.
• It uses memory cells to store weights of a neural network, which are crucial for performing computations.
• The row drivers and cell ADCs facilitate communication between the memory cells and the controller.
• The controller selects valid cell tiles and executes neural network-based arithmetic operations.
• The device redundantly stores weights of a specific layer in multiple valid cell tiles to ensure reliability and fault tolerance.

Potential Applications

• Artificial intelligence and machine learning: The neuromorphic device can be used in AI and ML applications that require efficient and parallel processing of neural networks.
• Robotics: It can be utilized in robotic systems that need real-time decision-making capabilities.
• Pattern recognition: The device can be employed in systems that analyze and recognize patterns in large datasets.
• Data processing: It can be used for high-speed data processing tasks, such as image and video processing.

Problems Solved

• Efficient computation: The device enables parallel processing of neural networks, leading to faster and more efficient computations.
• Reliability and fault tolerance: The redundant storage of weights in multiple cell tiles ensures that the system remains functional even if some tiles fail.
• Real-time decision-making: The neuromorphic device allows for quick decision-making, making it suitable for applications that require real-time processing.

Benefits

• Faster processing: The parallel processing capability of the device enables faster execution of neural network-based operations.
• Energy efficiency: The neuromorphic device is designed to minimize power consumption, making it energy-efficient.
• Scalability: The system can be scaled up by adding more cell tiles, allowing for larger and more complex neural networks.
• Fault tolerance: The redundant storage of weights ensures that the system remains operational even in the presence of failures.

Original Abstract Submitted

A neuromorphic device includes a plurality of cell tiles, each of the plurality of cell tiles including a cell array including a plurality of memory cells storing weights of a neural network, a row driver connected to the plurality of memory cells through a plurality of row lines, and cell analog-digital converters (ADCs) connected to the plurality of memory cells through a plurality of column lines, and a controller configured to select, form the plurality of cell tiles, a plurality of valid cell tiles storing the weights, execute a neural network-based arithmetic operation based on the plurality of valid cell tiles, and redundantly store weights of a first layer among a plurality of layers included in the neural network in a plurality of first valid cell tiles that are divided into a plurality of first tile groups.