Machine Learning Enhanced Analog-to-Digital Converters

Background

Data converters act as gatekeepers between analog and digital formats, translating information between the physical world and computing devices. Analog-to-digital converters (ADCs) specifically convert analog signals to digital numbers, where information can be processed, analyzed, and stored in digital format for human consumption. Many applications including communications, particle detectors, medical imaging, and scientific research require energy-efficient and high bandwidth ADCs. However, high-speed ADC design in advanced technology nodes still faces many challenges in order to be effectively scaled up.

Invention Description

Researchers at Arizona State University have developed a new class of high-performance ADCs that will use on-chip neural network to continuously correct static and dynamic errors generated from different mechanisms. This technology employs machine learning (ML) techniques, specifically neural networks, to enhance the performance of analog-to-digital converters (ADCs). The neural networks used can learn efficient representation without requiring explicit expressions. This method can simultaneously enhance the performance of ADCs and reduce power consumption of supply/reference voltage regulators. 

Potential Applications:

• 5G and high-speed communications
• Highly specific particle detectors
• Advanced scientific research
• Wearable healthcare technology

Benefits and Advantages:

• Suppresses ADC errors
• Does not require calibration through varying operating conditions
• High precision
• Increased energy-efficiency
• High-speed ADC design