Physics-informed neural networks (PINNs) provide strong generalization for battery state-of-health (SOH) estimation, but their sine activations make aggressive low-bit quantization and efficient edge deployment difficult. This paper presents a co-designed quantization and FPGA acceleration framework for PINN-based SOH estimation. We introduce per-channel GridWarp quantization-aware training, which learns neuron-wise scale and bias parameters to align the quantization grid with the non-uniform activation distribution. The learned transforms are compiled into per-channel activation LUTs that fuse sine evaluation, GridWarp, and requantization in hardware, while the fully connected layers are mapped to an output-neuron-parallel datapath with DSP packing and deferred offset correction. Across 11 evaluation settings, including six XJTU batches with different charge/discharge protocols, three TJU chemistry batches, and the MIT and HUST datasets, per-channel GridWarp reduces average W4A4 MAPE to 1.93%, compared with 5.51% for post-training quantization, 3.45% for QAT, and 2.41% for per-layer GridWarp. Implemented on a ZYNQ-7020 FPGA, the accelerator delivers 0.69 μs latency at 0.182W, with 37× and 80× lower latency than a desktop CPU and GPU, respectively.
Population aging and the prevalence of speech and upper-limb impairments leave many people for whom touch keyboards and voice input are hard to use. We assess the feasibility of deploying a Mandarin lip-reading model as a text-entry method that runs entirely on a consumer Android device. A compact deep model maps the lip-image sequence for a single character to a ranked list of pinyin, and a lightweight dictionary decoder assembles these predictions into Chinese text. The text is written into any application through a floating window and the Android accessibility service, and the model is deployed with TensorFlow Lite so that all computation stays on the device, protecting privacy. On a held-out split of a Mandarin dataset the model reaches about 88% syllable-level accuracy, and on a commodity smartphone the application runs within a practical memory budget with cross-application input. These results show that the complete pipeline is viable on current consumer hardware. Neither the problem nor the pipeline is specific to Mandarin; the same approach could be applied to other languages. Extending it beyond the speakers seen during training remains the main open challenge.
Over the last 15 years, our IEEE Consumer Technology Society members have been able to tune in to society and industry news as well as the latest advances in Consumer Technologies thanks to the IEEE Consumer Electronics Magazine.
