High Resolution Image Processing Using Multiple Floating Gate Nonvolatile Memory (MFGNVM)

Background:

Current computing systems based on CMOS technology and von Neumann architecture face significant challenges due to bottlenecks in data processing, limiting speed and power efficiency. Traditional flash memory devices typically handle only two data states, constraining the implementation of efficient multiply-accumulate (MAC) operations essential for AI and machine learning applications. These limitations have driven research toward memory technologies that can integrate data storage and computation for improved performance.

Technology Overview:

This technology introduces an innovative nonvolatile memory device utilizing multiple quantum wells (MQWs) to store multiple discrete states, going beyond the binary storage capability of conventional flash memory. This structure enhances data storage density and reliability through the use of high-bandgap barriers between quantum wells, effectively reducing charge leakage. The MQWNVM's design enables stable and reproducible multilevel storage, which is critical for the large-scale data processing needs of AI and neuromorphic computing. This integration of memory and processing addresses the inherent limitations of traditional computing architectures by enabling on-device MAC operations, thereby improving computational speed and energy efficiency. The compatibility of the fabrication process with standard CMOS technology ensures that this memory device can be adopted without significant manufacturing changes, facilitating practical deployment. Furthermore, theoretical models and circuit simulations confirm the device's effectiveness in neuromorphic circuits, particularly in vector-matrix multiplication tasks that underpin AI algorithms. Compared to resistive RAM and quantum dot gate NVM, this MQWNVM exhibits superior control over process variations and reproducibility, marking it as a robust solution for next-generation computing systems.

Advantages:

•    Multilevel data storage that enables greater information density compared to binary flash memory.
•    Reduced charge leakage and enhanced stability via high-bandgap barrier layers between quantum wells.
•    Integration of memory and processing capabilities, improving speed and power efficiency for AI tasks.
•    Compatibility with conventional CMOS fabrication processes, allowing easy adoption.
•    Improved reproducibility and reduced sensitivity to process variations relative to competing technologies.
•    Supports fundamental AI operations such as multiply-accumulate (MAC), facilitating neuromorphic computing implementations.

Applications:

•    Neuromorphic computing systems requiring efficient vector-matrix multiplication.
•    Artificial intelligence hardware accelerators that demand high-density, low-power memory solutions.
•    Advanced machine learning devices focusing on real-time and on-device data processing.
•    Next-generation nonvolatile memory in embedded systems where integration and energy efficiency are critical.
•    High-performance computing platforms seeking to overcome von Neumann bottlenecks through memory-compute integration.

Intellectual Property Summary:

Know-how based, patents available

Stage of Development:

TRL 2

Licensing Status:

This technology is available for licensing.