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.

The high theoretical energy densities of lithium-air batteries (LAB) make this technology an attractive energy storage system for future mobility applications. Li2O2growth process on the cathode relies on the surrounding chemical environment of electrolytes. Low conductivity and strong reactivity of Li2O2discharge products can cause overpotential and induce side reactions in LABs, respectively, eventually leading to poor cyclability. The capacity and reversibility of LABs are highly susceptible to the morphology of the Li2O2discharge products. Here, we identify for the first time that a seed layer formed by the combination of a cathode and an electrolyte determines the morphology of Li2O2discharge products. This seed layer led to its high reversibility with a large areal capacity (up to 10 mAh/cm2). Excellent OER (oxygen evolution reaction) was achieved by the formation of a favorable interface between the carbon electrode and electrolyte, minimizing the decomposition of the electrolyte. These remarkable improvements in LAB performance demonstrate critical progress toward advancing LAB into practical uses, which would exploit good reversibility of LABs in pouch-type cell arrangements with 1.34 Ah.

OBJECTIVE: Investigators from the Australo-Anglo-American Spondyloarthritis Consortium have reported additional genes associated with ankylosing spondylitis (AS) susceptibility including IL1R2, ANTXR2, and gene deserts at 2p15 and 21q22. We evaluated these new candidate genes in a large cohort of Korean patients with AS. METHODS: A group of 1164 patients with AS and 752 healthy controls were enrolled for our study. Eight single-nucleotide polymorphisms (SNP) were analyzed to define genetic association with AS by MassARRAY system. RESULTS: Significant positive associations of AS with endoplasmic reticulum aminopeptidase 1 SNP, rs27037 (p = 1.31 x 10–4), and rs27434 (p = 4.59 x 10–6), were observed. The rs10865331 of gene desert at 2p15 also showed a significant association with AS (p = 4.63 x 10–5). CONCLUSION: This is the first confirmation in a nonwhite population that genetic polymorphisms of rs27037, rs27434, and rs10865331 are associated with AS, implicating common pathogenetic mechanisms in Korean and white patients with AS.