Close
Close

Published

Qualcomm beefs up embedded SoC range with Snapdragon 820E

Against the backdrop of its ongoing buyout dispute with Broadcom, Qualcomm has unveiled the Snapdragon 820E, the latest addition to its embedded computing range of systems-on-chip. The new platform is aimed at premium devices and applications, which need more powerful hardware than its 400 and 600 Series.

Qualcomm is pitching the new platform at computer vision, AI and immersive multimedia developers, for virtual reality, digital signage, connected retail and robotics. The Snapdragon 820E houses a 64-bit ARMv8 quad-core CPU, the Qualcomm Kryo in this case, as well as an Adreno 530 GPU and a Hexagon 680 DSP. 802.11ac WiFi and Bluetooth 4.2 are also included, as well as satellite location support, for GNS, GLONASS, BeiDou, Gallileao, QZSS and SBAS.

Broadly, this is on par with Qualcomm’s 820 SoC, which was housed in smartphone flagships like the Samsung Galaxy S7, OnePlus 3, HTC 10, or Moto Z. But the 820 is not as powerful as the new Snapdragon 835, which features a smaller 10nm fabrication process with octa-core CPUs and the more powerful Adreno 540 GPU.

The 835 SoC was announced in November, and was swiftly followed by the 845, which houses the Neural Processing Engine (NPE) machine learning stack. However, Qualcomm has just announced the Qualcomm AI Engine, a suite of hardware and software components to provide AI capabilities on the Snapdragon 660, 820, 835, and 845.

Some of the first IoT devices to use the 820E are Allomind’s AR/VR glasses for in-flight entertainment, RetailNext’s Aurora v2 retail analytics sensor, and Nanolumens’ Aware digital display platform. Broadly, these are devices that require high computing power at the network edge, from developers looking to take advantage of the low-power consumption benefits of the ARM architecture, and Qualcomm’s pretty extensive developer community.

The embedded development kit is being launched initially through Arrow Electronics, in what looks like a ten-year deal. The DragonBoard 820C, compliant with the 96Boards open hardware specification (managed by Linaro), will let developers tinker with the embedded platform, thanks to those 96Boards-compliant modules and devices. It is shipping with Debian Linux and planned support for Linux OpenEmbedded. Qualcomm’s embedded range is listed as supporting Android and Windows 10 too.

The 820C system will support 3GB of LPDDR4 RAM, with 32GB of storage, according to the Arrow listing. For its video capabilities, important for machine vision and digital signage, the 820E will support 4k60 decode with 10-bit color over HDMI 2.0, with HEVC and VP9 support. For cameras, dual-capture support for 28MP and 13MP is cited, as well as support for dual-ISP 32MP capture (two 16MP camera inputs).

The new system slots in to Qualcomm’s current tier system, above the 410E and 600E embedded SoCs, which were launched back in September 2016. The new 820E is distinctly more powerful than the 600E, which houses a Krait 300 quad-core CPU and an Adreno 320 GPU.

Leon Farasati, director of product management at Qualcomm, explained that the lack of the Neural Processing Engine was an intentional choice, designed to offer the 820E as a general-purpose platform that can support as many potential applications as possible. Using the broad design ethos, Qualcomm hopes to allow developers to pick the AI-based approach that best suits their needs, and then build on top of the 820E platform.

On top of this, Farasati explained that the reason there are far fewer embedded SoCs than the mobile range is because of the Extended Life that Qualcomm needs to support. With the ten-year cycle, it is much easier for Qualcomm to support fewer designs, especially when all the necessary software is factored in – using Linux to create as universal a support option as possible, thanks to its community.

In addition, Qualcomm’s position as a core member of the Linaro project, which promotes Linux on ARM designs, is another key in this strategy, and Linaro’s 96Boards standard was chosen for the embedded series because of this. Ensuring access to that ecosystem is important for Qualcomm, in terms of developer adoption, and Qualcomm’s support burden is compounded further by all the work it has to do in its own supply chain, to ensure that its chips and modules are still available down the line.

Farasati added that there is a core difference in the approaches between Qualcomm’s main mobile line and its embedded ecosystem, in that the mobile lifecycle is much shorter and has fewer but much larger customers than the embedded side. The customer dynamic of having many smaller customers has led Qualcomm to use the broad-stroke approach to the embedded designs, and because it has fewer designs, it can pick the most appropriate features from the mobile developers too.

We pushed to find some sort of scale to evaluate the size of Qualcomm’s embedded wing, but Farasati said he could not provide numbers – for the amount of chips sold or staff deployed. Farasati said that Qualcomm has a dedicated engineering team that touches many other development teams in Qualcomm, and that it was a huge opportunity for Qualcomm – and that it was enjoying lots of momentum.

Qualcomm has also recently launched its Wireless Edge Services (WES) offering, which is designed to offer security and management capabilities for IoT deployments – tying in with ETSI’s Multi-access Edge Computing (MEC) standards vision.

Consisting of both cloud and on-chip software, Qualcomm’s WES will initially be launched on three chipsets – the industrial-focused MDM9206, the automotive MDM9628, and the QCA4020 smart home chipset. Eventually, WES will be extended to some of its Snapdragon SoC designs too.

The goal is to use the on-chip software to generate security keys, and then use the APIs to securely provide integrations with cloud applications. The first WES SDK will support Alibaba’s Cloud Link One, China Mobile’s OneNet, Gizwits, and Verizon’s ThingSpace. The launch does seem pretty China-centric, with Alibaba, Baidu, China Mobile, Gizwits, Mobike, and MeiG all voicing their support.

Also announced were new developer kits for the QCA4020 and QCA4024 SoCs, which combine WiFi, Bluetooth 5, and 802.15.4-based protocols like Zigbee and Thread. Both now come with pre-integrated support for HomeKit and the OCF’s IoTivity framework, as well as AWS IoT and Azure IoT support too.

Close