This paper proposes a novel technology of memory protection for the Non-Volatile Memory (NVM), applied to smart sensors and smart data. Based on the asymmetry of failure rate between the statuses of bit-0 and bit-1 in the non-volatile memory, as a result of the pollution of the radiation of cosmic ray, a two-dimensional parity with variable length error detection code (2D-VLEDC) for memory protection is proposed. 2D-VLEDC has the feature of variable length of redundant bits varied with content of data word in the NVM. The experimental results show that the same error detection quality could be achieved with a 30% redundancy improvement by applying the proposed 2D-VLEDC. The proposed design is particularly suitable for the use of safety-related fields, such as the automotive electronics and industrial non-volatile memories involved in the industrial automation.
The first-reported FocusNet display LSI integrating an 11-layer fully convolutional neural network and supporting color, contrast, and sharpness visual quality enhancement engine is designed. A FocusNet deep learning (DL) model is proposed to predict the human's focus region applying a tone mapping curve so as to decrease the pixel luminance, leading to 19.5mW of system power saving on OLED smart-phone. Implemented in 10nm CMOS technology, the proposed DL-assisted FocusNet display LSI costs 3.21M gates and 52kB SRAM with 0.7mm 2 area. Operated at 0.9V and 205MHz, this LSI consumes 8.54mW of power dissipation.