IoT on the Edge

Practical workloads on edge devices have a wide spectrum, from nW-scale, always-on sensor node to high performance deep-inference network on autonomous vehicle. To accommodate this, IME investigates heavily on silicon fabric and RISC-V based scalable SoC architecture.
IME’s Edge Computing group has developed a suite of design techniques to realize an ultra-low-power SoC. These SoC are critical for power constrained IoT applications such as machine health monitoring, neural signal processing and tire monitoring. We are also working on thousand-core Neuromorphic and Deep Learning hardware to enable new generation of pattern learning and recognition based real-world applications. These includes mobile or wearable devices, smart sensors, computers, robots and autonomous vehicles. 

Ubiquitous machine learning tasks on edge require unorthodox architectures to support enormous data flow and neural-net computation work loads. IME’s Edge Computing group deploys a holistic approach to enable true intelligent on edge: 
(1) Tailored PE/Neuron behavior for different computation tasks
(2) Novel memory device and In-memory computing circuits for bandwidth boosting 
(3) 2D and 3D large scale integration to minimize expensive off-chip communication
These techniques are developed within the framework of multiple multidisciplinary projects such as Neuromorphic Computing and Deep Learning Hardware.

The number of IoT devices is exploding at 17% CAGR towards 75B devices in 2025. Protecting the integrity and privacy of these IoT devices is a huge concern, not only for research community but also general public and the industry. To enable widespread deployment of IoT for smart nation, IME’ Edge Computing group is working on an end-to-end solution to safeguard users’ sensitive data. This is built on top of a trusted, intelligent silicon using SCA-resilient IPs, ultra-light weight ciphers and highly random entropy sources.

Applications that IME is developing are high resolution hybrid beam forming radars for automotive (ADAS Cars), drone applications and satellite radar applications. The MMICs are developed in CMOS, SiGe or III-V technology platforms depending on the application needs. Working with sub-system and DSP partners, mmwave radar is being customised with unique features for futuristic applications like contactless heart rate and breath rate monitoring, warehouse stock monitoring, non-destructive testing (NDT) applications  such as crack detection, imaging and many more. 

At IME we are developing MMIC solutions for last mile and front haul connectivity for cellular applications like 5G and 6G, and SOTM. Scalable phased arrays with indigenously developed MMICs with antenna in package (AiP) are being developed  for these applications bringing the whole form factor, cost down drastically while also reducing the reflective losses due to mmWave  transitions and dissipation due to the long feeding network to the antenna outside the chip.

IME has developed far field wirelessly powered IoT nodes and demonstrated in a real life network scenario with simultaneous operation of sensing nodes. Employing various circuit deign techniques, IME’s portfolio of low power radio transceiver SoCs for sensor networks spans broad spectrum of applications and standards. Demonstrated capabilities include, low-power wireless for 2.4GHz Zigbee-like sensing node for watch like compact IoT node; 400MHz ARIB T67 radio for applications such as smart metering, remote control and health monitors, ultra-wideband (UWB) and sub-GHz custom radios. The team also has developed RF digital radio transceiver IP blocks for TV white space (TVWS) Wi-Fi bands, for smart factory, smart nation applications.