Significant upgrades have been rolled out to Microchip Technology’s VectorBlox Accelerator software development kit (SDK) for convolutional neural network (CNN)-based artificial intelligence/machine learning (AI/ML) inference with PolarFire FPGAs. Along with the recent release of version 1.3 comes 19 new tutorials, bringing the total to 37. These include licence plate recognition in TensorFlow, MobileNet v1 and v2 in TensorFlow and TensorFlow2, various ResNet and SqueezeNet models in ONNX, torchvision ResNet in PyTorch and TinyYoloV4 in TensorFlow.
Network sizes have been reduced by 50% by optimising containers for weights and the built-in bit accuracy simulator’s speed has been tripled compared to version 1.2. Other dependencies that have been upgraded in the SDK include OpenVINO version 2021.4.1 and Python version 4.5.5.62. Microchip also upgraded the Libero software example project and programming file to support Libero software version 2021.2 and hot-swappable camera and HDMI inputs.
The company also added a new application notes section in GitHub that covers the following topics:
• Training custom Yolo V2/V3 networks via DarkNet.
• Running custom Yolo V2/V3 networks.
• Inspecting and converting TensorFlow 2 models.
• Face recognition: summary of the demo design.
• Face recognition: changing the faces database to identify other faces.
Powered by VectorBlox Accelerator SDK version 1.3, PolarFire FPGAs offer two to three times more power-efficient inference in mid-range FPGAs. This enables ML inference in applications that are thermally constrained and operates on battery power in environments that are susceptible to single-event upsets such as those experienced in low-orbit aerospace.
Converter power modules for 48 V networks Altron Arrow
Power Electronics / Power Management
The economic and quality-of-life benefits of electrification is driving the adoption of HV to 48 V DC-DC conversion across many markets with 48 V power modules becoming more common.
Read more...Memory for asset tracking Altron Arrow
DSP, Micros & Memory
The Page EEPROM, ST’s latest memory, has been designed for efficient datalogging and fast firmware upload/download in battery-operated devices.
Read more...Enhanced graphics on STM32U3
Design Automation
ST has introduced a new program designed specifically for users working with its mainstream STM32 MCUs, including STM32C0, U0, WBA, and U3 series.
Read more...Microchip and AVIVA Links collaboration Altron Arrow
News
Microchip and AVIVA Links have achieved groundbreaking ASA-ML interoperability, accelerating the shift to open standards for automotive connectivity.
Read more...Expanded STM32WL3x line for IoT sensors Altron Arrow
Telecoms, Datacoms, Wireless, IoT
The STM32WL31x and STM32WL30x are more tailored versions of the STM32WL33x for designers who wish to focus on specific features, while lowering their bill of materials.
Read more...Ultra-low jitter clock buffers Altron Arrow
Telecoms, Datacoms, Wireless, IoT
New SKY53510/80/40 family of clock fanout buffers from Skyworks are purpose-built for data centres, wireless networks, and PCIe Gen 7 applications.
Read more...Ultra-low power MEMS accelerometer Altron Arrow
Analogue, Mixed Signal, LSI
Analog Devices’ ADXL366 is an ultra-low power, 3-axis MEMS accelerometer that consumes only 0,96 µA at a 100 Hz output data rate and 191 nA when in motion-triggered wake-up mode.
Read more...Converting high voltages without a transformer Altron Arrow
Editor's Choice Power Electronics / Power Management
With appropriate power converter ICs, such as the LTC7897 from Analog Devices, many applications can be suitably powered without having to use complex and cost-intensive transformers.
While every effort has been made to ensure the accuracy of the information contained herein, the publisher and its agents cannot be held responsible for any errors contained, or any loss incurred as a result. Articles published do not necessarily reflect the views of the publishers. The editor reserves the right to alter or cut copy. Articles submitted are deemed to have been cleared for publication. Advertisements and company contact details are published as provided by the advertiser. Technews Publishing (Pty) Ltd cannot be held responsible for the accuracy or veracity of supplied material.
printer friendly version