Low-Light Image Enhancement Based on Nanoscale Stochastic Magnetic Tunnel Junctions

Area of Science:

• Image processing
• Hardware acceleration
• Non-volatile memory applications

Background:

• Low-light imaging challenges include poor luminance and lost details, limiting applications like surveillance.
• Conventional sigmoid-based enhancement methods suffer from latency and hardware compatibility issues.

Purpose of the Study:

• To develop a hardware-based low-light image enhancement scheme using magnetic tunnel junctions (MTJs).
• To overcome the limitations of conventional software-based sigmoid methods for improved luminance and detail clarity.

Main Methods:

• Images are converted to HSV color space for independent luminance adjustment.
• Magnetic tunnel junctions (MTJs) are utilized for hardware-based sigmoid function fitting via current-switching probability.
• Low pixel values are mapped to current amplitudes for efficient nonlinear luminance enhancement.

Main Results:

• The proposed MTJ-based method significantly enhances luminance and detail in low-light images.
• Achieved 3.82 dB improvement in PSNR and 9.91% in SSIM compared to the original sigmoid method.
• Demonstrated a 13.46% SSIM gain over a lightweight learning-based scheme.

Conclusions:

• The MTJ-based scheme provides efficient, hardware-accelerated low-light image enhancement.
• The approach is suitable for resource-constrained edge terminals and supports various resolutions.
• This work offers theoretical insights and engineering value for practical low-light image enhancement deployment.