Blog Archives

Efficient natural language processing on the edge is needed to interpret voice commands, which have become a standard way to interact with devices around us. Due to the tight power and compute constraints of edge devices, it is important to adapt the computation to the hardware conditions. We present a Transformer accelerator with a variable-depth adder tree to support different model dimensions, a SuperTransformer model from which Sub Transformers of various sizes can be sampled enabling adaptive model configuration, and a dedicated word elimination unit to prune redundant tokens. We achieve up to 6.9× scalability in network latency and energy between the largest and smallest Sub Transformers, under the same operating conditions. Word elimination can reduce network energy by 16%, with a 14.5% drop in F1 score. At 0.68V and 80MHz, processing a 32-length input with our custom 2-layer Transformer model for intent detection and slot filling takes 0.61ms and 1.6μJ.

Wireless communication enables an ingestible device to send sensor information and support external on-demand operation while in the gastrointestinal (GI) tract. However, it is challenging to maintain stable wireless communication with an ingestible device that travels inside the dynamic GI environment as this environment easily detunes the antenna and decreases the antenna gain. In this paper, we propose an air-gap based antenna solution to stabilize the antenna gain inside this dynamic environment. By surrounding a chip antenna with 1 ~ 2 mms of air, the antenna is isolated from the environment, recovering its antenna gain and the received signal strength by 12 dB or more according to our in vitro and in vivo evaluation in swine. The air gap makes margin for the high path loss, enabling stable wireless communication at 2.4 GHz that allows users to easily access their ingestible devices by using mobile devices with Bluetooth Low Energy (BLE). On the other hand, the data sent or received over the wireless medium is vulnerable to being eavesdropped on by nearby devices other than authorized users. Therefore, we also propose a lightweight security protocol. The proposed protocol is implemented in low energy without compromising the security level thanks to the base protocol of symmetric challenge-response and Speck, the cipher that is optimized for software implementation.

Distributed, mass-deployable mm-sized nodes with communication, sensing, and actuation capabilities such as microbots [1] and THz radios [2] are the key components of future collaborative large-scale networks with minimum intrusion. This vision is enabled by devices with miniaturization, low-cost fabrication, and low power. There is, therefore, a growing interest in mm-sized wake-up receivers (WuRxs) to save the limited battery energy of those devices. The size of RF WuRx is determined by the antenna, which is fundamentally proportional to the square of the carrier wavelength and is typically at cm2 level in the GHz range [3]. Using higher carrier frequency of 78GHz, the work in [4] demonstrates the previously smallest RF-WuRx size of 49mm2, but at the expense of high DC power of 25mW. Other modalities are also adopted for size reduction. In [5], an ultrasonic (US) WuRx is presented with a size of 14.5mm2, but requires an off-chip US transducer. In [6], an optical WuRx reduces the size to 0.85mm2 through integrated photodiodes, but the operation is susceptible to ambient light interference.