IoT Archives - Page 172 of 455 - SmartCitiesElectronics.com

Nvidia updates TAO toolkit to simplify AI modelling

Thu, Jun 09, 2022

The latest version of Nvidia’s Train, Adapt and Optimise (TAO) toolkit, is claimed to simplify and accelerate the creation of AI models for speech and vision AI applications.

Developers can use the power of transfer learning to create production-ready models customised and optimised for particular use cases without the need for massive amounts of data, explained Nvidia. Typical applications include detecting defects, translating languages, or managing traffic.

The latest version boosts developer productivity with the introduction of pre-trained vision and speech models. It also includes key new features such as ONNX model weights import, REST APIs, and TensorBoard integration.

It is now possible to build a new AI service or integrate AI into an existing one with REST architecture (representational state transfer) application programming interfaces (APIs). The TAO Toolkit service can be managed and orchestrated on Kubernetes. With TAO Toolkit as-a-service, IT managers can deliver scalable services using industry-standard APIs.

It is also possible to fine-tune and optimise non-TAO models with TAO. It is possible to import pre-trained weights from ONNX and take advantage of TAO features like pruning and quantisation, with support for image classification and segmentation tasks.

Developers can understand model training performance by visualising scalars such as training and validation loss, model weights, and predicted images in TensorBoard. They can also compare results between experiments by changing hyperparameters.

In addition, pre-trained models speed up the customisation process with the ability to fine-tune through the power of transfer learning, with less data. Some of the new pretrained models in this latest version can apply data gathered from lidar sensors for robotics and automotive applications, advised Nvidia. It is also possible to classify human actions based on human poses. This can find applications in public safety, retail, and worker safety use cases.

Another use for pre-trained models is to estimate key points on humans, animals, and objects to help portray actions or simply define the object shape.

Users can create custom voices with just 30 minutes of recorded data to power smart devices, game characters, and quick service restaurants for example.

Enterprise support for TAO Toolkit is available with Nvidia AI Enterprise, an end-to-end software suite for AI development and deployment.

http://www.nvidia.com

> Read More

Sitara AM62 processors expand edge AI while cutting the power budget

Tue, Jun 07, 2022

Sitara AM62 processors have been released by Texas Instruments. The company said that they help expand edge artificial intelligence (AI) processing and the low power design enables support for dual-screen displays and small size human machine interface (HMI) applications.

According to Texas Instruments (TI), the next generation of HMI will bring new ways of interacting with machines, such as enabling gesture recognition to give commands in a noisy factory environment or enabling the control of machines by phones or tablets with a wireless connection. The AM62 processors bring analytics to edge devices at low power, including suspend states as low as 7mW. There is also no need to design for thermal considerations, said TI, which can give engineers flexibility to deploy new capabilities in size-constrained applications or industrial environments. Adding edge AI features to HMI applications, including machine vision, analytics and predictive maintenance, can advance HMI beyond a mere interface to enable human-machine interaction.

The AM62 processor starts at less than $5.00 to bring low-cost analytics to HMI devices with basic camera-based image processing and edge AI functions, such as detecting and recognising objects, said TI. AM62 processors also enable dual-screen, full HD displays and support for multiple operating systems, including Mainline Linux and Android operating systems. AM62 processors also support both wired and wireless connectivity interfaces.

Additionally, the AM62 processors can reduce power consumption in industrial applications by as much as 50 per cent compared to competing devices, said TI. They can thus enable an application powered by AA batteries to remain on for over 1,000 hours. This is possible, said TI, due to a simplified power architecture. The device features only two dedicated power rails and five power modes. Deep sleep mode at less than 5mW enables longer battery life, while an active power of less than 1.5W is made possible by a core voltage of 0.75V. Reducing system power consumption extends battery life and helps engineers meet design requirements for handheld or size-constrained devices. Achieving optimal power performance is further simplified with the new TPS65219, a companion PMIC specifically designed to meet AM62 processor power supply requirements, said TI.

There is a wide range of tools and resources for AM62 processors including multiple open-source software such as Mainline Linux, to simplify the application development process and help accelerate time to market. A hardware ecosystem includes a third-party evaluation module.

The AM625 and AM623 processors are now available in a 13 x 13mm, 425-pin ALW package. It is available directly from TI and authorised distributors.

TI will showcase the AM62 processors and demonstrate system-level solutions for edge AI and electric vehicle charging HMI applications in at Embedded World in Nuremberg, Germany (21 to 23 June 2022) at its stand in Hall 3A #215.

http://www.TI.com

> Read More

Radio wave ranging sensor is sensitive enough for gesture initiation

Mon, Jun 06, 2022

Radio wave ranging sensors announced by Socionext are able to detect minor movements in positional information for tracking. The SC1240 series radio wave ranging sensors use the 60GHz band and have a built-in signal processing circuit for detecting the position and movement of a person.

The sensor that complies with the global broadband 60GHz radio equipment standard. It is able to detect the position and movements of a person, said Socionext and its high precision sensing and built-in signal processing circuits use a 6.8GHz (57.1 ~ 63.9GHz) wide band to detect minor movements as positional information, enabling it to be used in advanced applications such as tracking human movements and operation by gestures.

The sensor has built-in range finding and angle calculating signal processing circuitry, believed to be the first in the industry. It is based on the company’s mmwave radio communication LSIs and 24GHz radio wave range finding sensors.

The integrated SC1240 series uses 3D position information without the need for high frequency or advanced signal processing technology, claimed Socionext. It includes multiple antennas, wireless circuitry, ADCs, FIFO memory, SPI (serial peripheral interface), a sequencer that flexibly changes the duty cycle and controls power consumption, and a signal processing circuit for distance measurement and angle calculation. It also provides an autonomous activation function. These features allow users to easily obtain the height and positional data on the X, Y and Z planes for precision sensing actions such as the detection of multiple people or gesture without contact.

Average power consumption is 0.72mW at 0.1 per cent duty cycle operation, reported Socionext. The sensor is supplied in an FC-BGA measuring 4.0 x 7.0 x 1.2mm.

Sample shipments are scheduled for Q2 2022, and mass production in Q1 2023.

Socionext supplies SoCs for use in automotive, data centre, networking and smart devices.

Socionext Inc. is headquartered in Yokohama, and has offices in Japan, Asia, the United States and Europe.

Socionext Europe has headquarters is in Frankfurt, Germany with offices in Munich, Germany and Maidenhead, UK. The Munich Design and Support Centre specialises in embedded graphic, hardware and software product development for the automotive sector. The Maidenhead office in the UK supports Socionext’s IP Development & Engineering Centre with a focus on high speed data conversion IP. Socionext Europe offers customers a range of custom SoCs.

http://www.socionext.com

> Read More

Solid State Supplies adds Silicon Labs’ BG24 and MG25 SoCs

Wed, Jun 01, 2022

Distributor Solid State Supplies can now supply Silicon Labs’ BG24 and MG24 series of 2.4GHz wireless SoCs for Bluetooth and multiple-protocol operations, respectively.

The low power BG24 and MG24 series support multiple wireless protocols and incorporate PSA Level 3 Secure Vault protection, making the SoCs suitable for smart home, medical and industrial applications.

The 2.4GHz wireless SoCs are claimed to have the industry’s first integrated AI/ML (artificial intelligence / machine learning) accelerators, support for Matter, Zigbee, OpenThread, Bluetooth Low Energy, Bluetooth mesh, proprietary and multi-protocol operation. In addition they have the highest level of industry security certification, said Solid State Supplies, together with low power capabilities and the largest memory and flash capacity in the Silicon Labs portfolio.

There is a new software toolkit designed to allow developers to quickly build and deploy AI and ML algorithms using some of the most popular tool suites like TensorFlow.

Equipped with built-in dedicated AI/ML accelerators (specialised hardware designed to handle calculations in AI/ML implementations), the BG24 and MG24 “represent a huge step forward in the quest to bring the potential of AI and ML to life in real world applications. Their impressive battery life will help to bring AI/ML applications and wireless high performance to battery-powered edge devices,” commented Matt Cook, product group manager at Solid State Supplies.

The single-die BG24 and MG24 SoCs combine a 78MHz Arm Cortex-M33 processor, 2.4GHz radio, 20bit ADC, an optimised combination of flash (up to 1536kB) and RAM (up to 256kB), and an AI/ML hardware accelerator for processing ML algorithms while offloading the Arm Cortex-M33, so applications have more cycles to do other work. They support a range of 2.4GHz wireless IoT protocols. System on Module (SoM) versions will also be available to support lower volume applications.

PSA Level 3-Certified Secure Vault, the highest level of security certification for IoT devices, provides the security needed in products like door locks, medical equipment, and other sensitive deployments at risk from external threats.

The SoCs and development kits are available now. SoM versions are due for release later in 2022.

In addition to natively supporting TensorFlow, Silicon Labs has partnered with AI and ML tools providers, SensiML and Edge Impulse, to ensure that developers have an end-to-end toolchain that simplifies the development of ML models optimised for embedded deployments of wireless applications. Using this AI/ML toolchain with Silicon Labs’s Simplicity Studio and the BG24 and MG24 SoCs, developers can create applications that draw information from various connected devices, all communicating with each other using Matter to then make intelligent machine learning-driven decisions.

http://www.sssltd.com

> Read More

Nvidia updates TAO toolkit to simplify AI modelling

Sitara AM62 processors expand edge AI while cutting the power budget

Radio wave ranging sensor is sensitive enough for gesture initiation

Solid State Supplies adds Silicon Labs’ BG24 and MG25 SoCs

About Smart Cities