LG's new edge AI SoC debuts on Ubuntu dev kit

LG’s new edge AI SoC debuts on Ubuntu dev kit

Jan 8, 2021 — by Eric Brown

— 225 views

LG unveiled a quad -A53 “LG8111 AI SoC” with an “LNE” NPU. The LG8111 is available on an Ubuntu powered “Eris” dev kit designed to work with AWS Greengrass.

LG’s IoT-focused ThinQ division has launched a Linux-driven dev kit that showcases a new LG8111 AI SoC equipped with an LG Neural Engine (LNE). The LG8111 Development Kit, code-named Eris, is focused on local edge AI applications using the cloud-connected AWS Greengrass platform.

LG8111 AI SoC (left) and Eris development kit
(click images to enlarge)

Major features include space/situation and near/far distance recognition, user and object location, and motion detection. The platform also supports voice control with keyword/synonym recognition, as well as sensor fusion.

The quad-core, Cortex-A53 LG8111 AI SoC is clocked to 1GHz and has a 1MB L2 cache. The SoC integrates the homegrown LNE chip, which offers an inference/learning engine and an extensible architecture. No TOPS rating was provided for the NPU, which supports TensorFlow, TensorFlow Lite, and Caffe.

The LG8111 AI SoC is further equipped with ISPs for driving dual 5-megapixel MIPI-CSI2 or Parallel cameras. The image processors support lens distortion correction and noise reduction. There is also a video encoder with full HD encoding and a video accelerator with dedicated vision DSP and features like feature extraction, image warping, and SFC.

A voice/audio engine includes a DSP-enabled voice activity detector and standby voice recognition. Other features include a sensor fusion and analysis with detection of temperature, vibration, and odor.

A crypto engine and TrustZone support is available along with integrated dual-band 802.11 b/g/n support. The SoC supports up to 2x USB 2.0 host ports plus I/O including SPI, UART, PWM, I2C, SDIO, and GPIO.

Conceptual diagram showing LG8111 with AWS Greengrass (left) and Eris detail view (caption: 1) USB host, 2) micro-USB debug, 3) USB Type-C for power, 4)reset button
(click images to enlarge)

The roughly Raspberry Pi sized Eris dev kit is equipped with a DQ1-B0 SOM v1.5 module that incorporates the LG8111 AI SoC running Ubuntu 18.04. The module has an undisclosed amount of DDR3 RAM and eMMC. The carrier board provides USB host, micro-USB debug, and USB Type-C power ports, as well as a reset button.

We also see dual mic interfaces, a camera interface, a second micro-USB port, some GPIO pins, and JTAG. The CNXSoft post that alerted us to the product lists a battery, -40 to 85°C support, and 2.4GHz WiFi (perhaps on the module).

LG provides documentation for connecting the board to AWS Greengrass, which provides for autonomous, offline operation of AWS IoT-connected devices. AWS Greengrass enables embedded devices to respond quickly to local events and operate with intermittent connections while reducing the latency and cost of transmitting IoT data to the cloud. Because AWS Greengrass uses AWS Lambda programming models, customers can develop and test machine learning software in the cloud before deploying it to devices.

It also appears the Eris device can connect to LG’s ThinQ.Cloud platform using a RESTful API. The last time we heard from LG was in 2019 when the Korean tech giant announced a collaboration with Qt to integrate LG’s Linux-based webOS Open Source Edition with the Qt development platform for automotive, smart home, and robotics applications.

Further information

No pricing or availability information was provided for the
LG8111 AI SoC and Eris LG8111 Development Kit. More information may be found at LG’s LG8111 AI SoC product page (translated) and GitHub wiki (PDF).