Heterojunction-Driven Stochasticity: Bi-Heterojunction Noise-Enhanced Negative Transconductance Transistor in Image Generation

Abstract

Reliable true-random number generator (TRNG) hardware demands amplified intrinsic noise and multi-bit entropy output, which are difficult to achieve in conventional single-device TRNG implementation. A bi-heterojunction noise-enhanced negative transconductance (BHN-NTC) transistor is presented, incorporating an asymmetric PTCDI-C13 layer into an NTC transistor. This design enhances electron injection, expanding the NTC region (19 → 27 V) and increasing negative transconductance (−0.036 µS at VGS = −11 V → −0.073 µS at VGS = −15 V) by reducing the electron injection barrier (≈2.13 eV → ≈0.41 eV). The bi-heterojunction configuration introduces a strong correlation between noises, including trapping/detrapping and generation/recombination processes. This property enables a threefold higher entropy throughput in TRNG, achieving a 3-bit output per sampling event. The BHN-NTC-driven TRNG leverages increased noise-induced entropy to generate more diverse latent vectors, mitigating mode collapse and enabling the synthesis of high-quality, realistic images. This significantly enhances StyleGAN2-based image generation, improving performance metrics such as Frechet inception distance (FID) (18.7 → 8.3), kernel inception distance (KID) (0.024 → 0.009), inception score (IS) (6.5 → 9.2), and multi-scale structural similarity (MS-SSIM) (0.43 → 0.21). Consequently, the BHN-NTC transistor establishes a scalable stochastic noise platform, advancing applications in secure electronics and probabilistic stochastic computing.