Upscaling layers are important components of modern deep learning networks but often pose computational challenges for hardware (HW) accelerators. This article addresses this issue by introducing a novel layer-replacement technique to efficiently process upscaling layers using existing hardware-supported operations like depthwise convolutions and maximum pooling. To minimize the number of replacement layers, we propose an efficient layer number reduction algorithm. Experimental results on four deep neural networks demonstrate a significant speedup ranging from$1.58\times $to$32.88\times $compared to the original HW/software (SW) execution approach, and from$3.65\times $to$19.21\times $compared to the software-only solution, with minimal hardware overhead (0.068% more field-programmable gate array (FPGA) look-up tables (LUTs)). Notably, our technique introduces no numerical errors and maintains comparable input data processing quality to the original network.

In this paper are given two novel algorithms for minimization of recursive Boolean formula (RBF), which is adequate for implementation of N-input 1-output Boolean functions (BFs) over basis {imply, false} using two memristors. Both of our algorithms are direct consequence of necessary and sufficient conditions related to regular ordering of positive product terms within recursive formula. The results demonstrate how developed algorithms provide up to 26% gain in average number of implications and shorter recursive Boolean formula length in up to 77% of problem instances than previously published algorithms, tested on the set of all 4-input 1-output BFs.