Search

Transformer models are difficult to employ in mobile devices due to their memory- and computation-intensive properties. Accordingly, there is ongoing research on various methods for compressing transformer models, such as pruning and quantization. However, general computing platforms, such as central processing units (CPUs) and graphics processing units (GPUs), are not energy-efficient to accelerate the pruned model because the unstructured sparsity they exhibit causes degradation of parallelism. In this article, we propose a low-power accelerator for transformers that can handle various levels of structured sparsity induced by line pruning with different granularity. Our approach accelerates pruned transformers in a head-wise and line-wise manner. We present a head reorganization and shuffling method that supports head-wise skip operations and resolves the load imbalance problem among processing engines (PEs) caused by the varying number of operations in each head. Furthermore, we implemented a sparse quantized general matrix-to-matrix multiplication (SQ-GEMM) module that supports line-wise skipping and on-the-fly tile-based dynamic quantization of activations. As a result, compared to mobile GPU and CPU, the proposed accelerator improved the energy efficiency by $2.9\times $ and $12.3\times $ for the detection transformer (DETR), and $3.0\times $ and $12.4\times $ for the vision transformer (ViT) models, respectively. In addition, our proposed mobile accelerator achieved the highest-energy efficiency among the current state-of-the-art FPGA-based transformer accelerators.

Useful thermal bridge performance indicators (ITBs) of existing buildings may differ from calculated thermal bridge performance derived theoretically due to actual construction conditions, such as irregular shapes and aging. To fill this gap practically, a more realistic quantitative evaluation of thermal bridge on-site needs to be considered to identify thermal behaviors throughout exterior walls and thus improve the overall insulation performance of buildings. In this study, a case study is conducted using an infrared thermal imaging method to evaluate the thermal bridge of an existing building practically. The study's main purpose is to review the thermal bridge performance indicators measured by the steady-state model under field conditions in terms of convergence and uncertainty. Bayesian Markov Chain Monte Carlo (MCMC) is used to examine the validity of the results by deriving evaluation results in the form of distribution, including uncertainty. After the measurement was completed, an analysis of the results was conducted. As a result of measurement for 3 days, it was found that the thermal bridge part had 1.221 times more heat loss than the non-thermal bridge part, which showed a 6.7% deviation from the numerical method. However, the uncertainty was 0.225 (18.4%, CI 95%), a large figure. This is due to the influence of field conditions and shows the limitations of the steady-state measurement model. Regarding the convergence of the results, the measurement results were found to converge continuously as the measurement time increased. This suggests that valid results can be obtained in the field if the measurement is performed for a sufficient time, even with a thermal bridge measurement method assuming a steady-state.

Battery systems are critical factors in the effective use of renewable energy systems because the self-production of electricity by renewables for self-consumption has become profitable for building applications. This study investigates the appropriate capacity of the battery energy storage system (BESS) installed in all-electric zero-energy powerhouses (AEZEPHs). The AEZEPH used for this study is a highly energy-efficient house. Its criteria indicate that all the electrical energy within the home is covered based on the generated electricity from onsite renewable energy systems, including that the annual net site energy use is almost equal to zero. The experiment for measured data of electricity consumed and generated in the buildings is conducted for a year (i.e., January through December 2014). Based on the measured data, patterns of the electricity consumed by the AEZEPH and generated by an onsite renewable energy system (i.e., photovoltaic (PV) system) and BESS's appropriate capacity are then analyzed and evaluated using the electric energy storage (EES) analysis tool, named Poly-sun. This study indicates that self-consumption can be increased up to 66% when the ESS system is installed and used during operating hours of the PV system. The amount of received electricity during the week tends to be reduced by about two times.