Microchip reduces power needs for AI functions

Compute power for artificial intelligence (AI) inference at the edge is reduced, says Microchip, when using the memBrain neuromorphic memory technology.

AI processing is moving from the cloud to the edge of the network, meaning that battery powered and deeply embedded devices are challenged to perform AI functions, such as computer vision and voice recognition.

Via its subsidiary, Silicon Storage Technology (SST), Microchip offers the memBrain which is based on SuperFlash technology and optimised to perform vector matrix multiplication (VMM) for neural networks. The memory technology is claimed to improve system architecture implementation of VMM through an analogue in-memory compute approach, which, in turn, is claimed to enhance AI inference at the edge.

Current neural net models may require 50M or more synapses (weights) for processing which can challenge off-chip DRAM bandwidths, causing a bottleneck for neural net computing and an increase in overall compute power. The memBrain stores synaptic weights in the on-chip floating gate, which significantly improves system latency, says Microchip. Compared to traditional digital DSP and SRAM/DRAM based approaches, it delivers 10 to 20 times lower power and “significantly reduced overall bill of materials (BOM)”.

The memBrain is expected to advance machine learning (ML) capacities in edge devices, with reduced power in memory compute operations in AI.

SST will present Microchip’s memBrain product tile array-based architecture at the AI/ML session track on flash performance scaling at the 2019 Flash Memory Summit this week (6 to 8 August 2019) at the Santa Clara Convention Center in Santa Clara, California, USA.

SST offers design services for memBrain and SuperFlash technology and a software toolkit for neural network model analysis.

Silicon Storage Technology provides embedded flash technology in addition to developing, designing, licensing and markets proprietary and patented SuperFlash memory technology for the consumer, industrial, automotive and IoT markets.

SST was founded in 1989 and acquired by Microchip in April 2010 and is now a wholly owned subsidiary of Microchip.

http://www.microchip.com

> Read More

Microchip presents serial memory controller for AI and ML

Microchip has entered the memory infrastructure market, offering what it claims to be the first commercially serial memory controller for artificial intelligence (AI) and machine learning (ML).

The SMC 1000 8x25G enables four times the memory channels of parallel-attached DDR4 DRAM and low latency, says Microchip. It has been introduced as the computational demands of AI and ML workloads accelerate, highlighting the shortfall in DRAM, which require an increased number of memory channels to deliver more memory bandwidth.

The SMC 1000 8x25G enables CPUs and other compute-centric SoCs to use four times the memory channels of parallel attached DDR4 DRAM within the same package footprint. This enables serial memory controllers to deliver higher memory bandwidth and media independence to these compute-intensive platforms with low latency.

As the number of processing cores within CPUs has risen, the average memory bandwidth available to each processing core has decreased because CPU and SoC devices cannot scale the number of parallel DDR interfaces on a single chip to meet the needs of the increasing core count. The SMC 1000 8x25G interfaces to the CPU via 8-bit Open Memory Interface (OMI)-compliant 25Gbits per second lanes and bridges to memory via a 72-bit DDR4 3200 interface. This formation reduces the required number of host CPU or SoC pins per DDR4 memory channel “significantly”, says Microchip, allowing for more memory channels and increased memory bandwidth.

The SMC 1000 8x25G is the first memory infrastructure product in Microchip’s portfolio that enables the media-independent OMI.

Data centre application workloads require OMI-based differential dual-inline memory modules (DDIMMs) to deliver the same high-performance bandwidth and low latency results of today’s parallel-DDR based memory products. The low latency of the SMC 1000 8x25G delivers less than four nanoseconds incremental latency to the first DRAM data access and identical subsequent data access performance, reports Microchip. OMI-based DDIMM products have virtually identical bandwidth and latency performance to comparable LRDIMM products, concludes the company.

A CPU or SoC with OMI support can use various types of media, allowing designers to select appropriate cost, power and performance metrics without having to integrate a unique memory controller for each type. Microchip points out that CPU and SoC memory interfaces today are typically locked to specific DDR interface protocols, such as DDR4, at specific interface rates.

SMART Modular, Micron and Samsung Electronics are building multiple pin-efficient 84-pin DDR4 DDIMMs with capacities ranging from 16 to 256Gbyte. These DDIMMs will leverage the SMC 1000 8x25G and will seamlessly plug into any OMI-compliant 25Gbits per second interface, Microchip advises.

The SMC 1000 is supplied with ChipLink diagnostic tools that provide extensive debug, diagnostics, configuration and analysts tools with an intuitive graphics user interface (GUI).

The SMC 1000 8x25G is sampling now.

http://www.microchip.com

> Read More

Automation engine manages IoT across countries and networks

Arm has announced Pelion Connectivity Management 2.0, an automation engine to scale the IoT and automatically manage millions of devices across multiple countries and different networks.

The company believes that a single platform for managing IoT connectivity across all a mobile network operator’s (MNOs) devices’ lifecycles. Arm says MNOs can drive additional revenue opportunities by enabling their enterprise clients the capability to automatically configure IoT devices. The automation capabilities are claimed to significantly improve the IoT services, speed and reliability of connectivity, which can be the provisioning of new devices or ensuring monthly bills never go over a set threshold.

Automation capabilities include real-time data triggers to create rules that are triggered by what is happening in the physical world.

The seamless remote subscriber identity module (eSIM) provisioning management ensures the correct service and commercial configuration and IoT deployment operational costs can be reduced, adds Arm.

The company believes Pelion Connectivity Management 2.0 will help enterprises overcome the challenges of moving from pilot IoT projects to a global scale, which enterprises have not been able to achieve before as manually managing every device across multiple countries and different mobile networks had not been possible until now.

“With our enterprise customers increasingly expanding internationally, as well as the rise of new technologies such as NB-IoT and eSIM, we wanted an IoT connectivity management service that can provide a single pane of glass regardless of underlying technology, MNO or geography,” said Carlos Lourenço, director corporate IoT business unit, of Portugal’s mobile network operator, NOS.

Michele Mackenzie, principal analyst, Analysys Mason, believes this introduction will accelerate IoT adoption. “Network operators and enterprise customers might start with small estates of devices but look to scale up quickly as the benefits of IoT are realised,” she said. “For millions (and potentially billions) of devices to be connected to telecoms networks in a timely, cost efficient, scalable way platform services will need to become increasingly automated.”

http://www.arm.com

> Read More

32-bit microcontrollers pack functions into less space

Reducing PCB footprint by up to 59 per cent, the RX651 32-bit microcontrollers provide functionality for space constrained IoT connectivity modules and edge devices.

Renesas Electronics announced the four new RX651 microcontrollers, supplied in small LQFP and BGA packages, that have been added to the RX651 family. The 64-pin BGA package measures 4.5 x 4.5mm, which reduces footprint size by 59 per cent compared to the 100-pin LGA and a 64-pin LQFP, measuring 10 x 10mm reduces the footprint by 49 per cent compared with the 100-pin LQFP. The microcontrollers are designed for a secure endpoint devices employing compact sensor and communication modules in industrial, network control, building automation, and smart metering systems operating at the IoT edge.

The IoT and Industry 4.0 both rely on higher performance, smaller form factor connectivity modules that support confidentiality, data integrity and availability in connected devices. According to Daryl Khoo, vice president of marketing, IoT Platform business division, Renesas Electronics said: “The 64-pin RX651 MCUs give customers the small footprint, high performance, and security features they need to safeguard their connected industrial and manufacturing systems against cyber-attacks.”

The RX651 MCUs integrate connectivity, Trusted Secure IP (TSIP), and trusted flash area protection that enable flash firmware updates in the field through secure network communications. The increase in endpoint devices operating at the edge has increased the need for secure over-the-air (OTA) firmware updates. These RX651 devices support this reprogramming requirement with integrated TSIP and enhanced flash protection. Enhanced security features are based on the RXv2 core and 40nm process that provide a 520 CoreMark score at 120MHz, and strong power efficiency with a 35 CoreMark/mA score as measured by EEMBC Benchmarks.

The integrated dual bank flash memory enables engineers to realise high root-of-trust levels through a combination of TSIP that protects the encryption key. Encryption hardware accelerators including AES, 3DES, RSA, SHA, and TRNG and code flash area protection to safeguard boot code from reprogramming. The dual bank flash function supports both background operation) and SWaP, making it easier for manufacturers to execute in-the-field firmware updates securely and reliably.

The RX651 microcontrollers monitor the operating state of machinery from both inside and outside the factory, enabling data exchanges to change production instructions, and reprogram microcontroller memory to update equipment settings.

The microcontrollers are available now from Renesas Electronics’ distributors. Target boards and starter kits, combined with the e2 studio integrated development environment (IDE) are also available.

http://www.renesas.com

> Read More

About Smart Cities

This news story is brought to you by smartcitieselectronics.com, the specialist site dedicated to delivering information about what’s new in the Smart City Electronics industry, with daily news updates, new products and industry news. To stay up-to-date, register to receive our weekly newsletters and keep yourself informed on the latest technology news and new products from around the globe. Simply click this link to register here: Smart Cities Registration