News

Sense, Process, Connect: Synaptics Launches SoCs and Dev Kit for Edge AI

April 19, 2024 by Jake Hertz

Synaptics' new SoCs and dev kit bring "out-of-the-box" embedded AI to any product, independent of the data center.

While Synaptics has traditionally focused on a few processor verticals, the company branched out to bring advanced AI to edge IoT devices at Embedded World. There, the company announced its Synaptics Astra platform, including its SL-Series of AI-native IoT processors. The company also released its Astra Machina Foundation Series development kit to support the new platform. 

Describing the new edge-AI-focused offerings, Synaptics president and CEO Michael Hurlston remarked that the new family of embedded processors hits on the three fundamental elements of an IoT system: sense, process, and connect. 

 

Siddarth Chandrasekar

Siddarth Chandrasekar, Synaptics' senior director of product marketing for multimedia processors.

 

Synaptics claims that the new SL-Series can help embedded designers integrate AI directly into devices without sending any data to data centers, absolving latency and security concerns associated with cloud computing. 

 

SL1680: Tailored for Versatility 

The first of Synaptics’ offerings is the SL1680, a versatile SoC designed for multi-modal consumer, enterprise, and industrial IoT applications. 

Key features of the SL1680 include a quad-core Arm Cortex-A73 processor that can reach up to 2.1 GHz per core, delivering a significant computational power of up to 40,000 DMIPS. This is supplemented by a dedicated neural processing unit (NPU) that achieves more than 7.9 tera operations per second (TOPS), making it highly effective for localized machine learning applications.

The SoC also supports a broad range of video and audio codecs and formats, boasting capabilities for 4K video encoding and decoding at 60 frames per second. It supports advanced video codecs like AV1, H.265, and VP9, among others. These multimedia features are enhanced by the inclusion of Imagination Technology’s PowerVR Series9XE GE9920 GPU, which supports a wide array of graphical and compute APIs and can deliver up to 200 million polygons per second, with a pixel fill rate of 5.6 Gpixels/s.

 

High-level block diagram of the SL1680

High-level block diagram of the SL1680. Image used courtesy of Synaptics
 

On the connectivity front, the SL1680 offers comprehensive options, including dual-channel PCI Express 2.0, Gigabit Ethernet MAC, and advanced wireless standards through Synaptics’ compatible wireless connectivity portfolio. Memory interfacing is robust as well, with support for up to 4 GB of LPDDR4/LPDDR4x memory, enhancing the SoC’s ability to handle complex tasks and large datasets efficiently.

The SL1680 incorporates a suite of security features such as a secure boot, a dedicated security processor (Arm Cortex-M3), and a Direct Rendering Manager (DRM) engine, ensuring secure operations and data protection across applications. Designed with an emphasis on performance-per-watt efficiency, the SL1680 may enable cost-optimized systems that do not sacrifice capability for power economy.

 

SL1640: Optimized for Cost and Power

Synaptics developed its SL1640 SoC for IoT applications with a design focus on cost and power optimizations. 

At the heart of the SL1640 is a quad-core Arm Cortex-A55 CPU cluster that can reach up to 1.9 GHz for each core. This setup provides a substantial processing backbone, achieving up to 24K DMIPS while supporting dynamic voltage and frequency scaling to enhance power efficiency. The CPU is complemented by 32 KB of I-cache and 32 KB of D-cache per core, with an additional 64 KB per core of L2 cache and 512 KB of shared L3 cache.

 

High-level block diagram of the SL1640

High-level block diagram of the SL1640. Image used courtesy of Synaptics
 

For AI tasks, Synaptics equipped the  SL1640 with a multi-TOPS NPU optimized for edge AI inferencing. This is supported by TensorFlow Lite and Android NN API, allowing the SoC to handle complex neural network models natively. Additionally, the Imagination PowerVR Series9XE GE9608 GPU facilitates advanced 3D graphics and AI acceleration, further enhancing the chip's utility in graphics-intensive applications.

The SL1640 also features a broad range of connectivity options, including high-speed USB, PCIe, and advanced options for Ethernet and wireless communication. It offers up to 4 GB of DDR4 or LPDDR4x memory and includes a comprehensive set of interfaces for I/O, storage, and peripheral connectivity. This includes support for high-speed storage interfaces like eMMC 5.1, enhancing its suitability for complex, data-intensive IoT devices.

 

SL1620: Built for Advanced Graphics and AI Acceleration

The final processor Synaptics announced at Embedded World is the SL1620 IoT-focused SoC.

The processor is powered by a quad-core Arm Cortex-A55 CPU, with each core capable of running up to 1.9 GHz. The SoC supports dynamic voltage and frequency scaling along with power gating to manage power. It also includes a dual-core Imagination Technology BXE-2-32 GPU that operates up to 800 MHz, with a range of APIs for advanced graphics and AI acceleration.

 

High-level block diagram of the SL1620

High-level block diagram of the SL1620. Image used courtesy of Synaptics
 

The SL1620 is rife with security features, with secure boot using RSA digital signature verification and a DRM engine compatible with multiple encryption standards, including AES and RSA. Additional security features include a true random number generator and secure control based on Arm TrustZone, along with customizable JTAG authentication options.

The SL1620's connectivity features include an integrated 802.3 GMAC for Ethernet for RGMII and RMII interfaces with VLAN-tagging capabilities. The processor also offers significant peripheral support, including one USB 3.0 host, one USB 2.0 on-the-go (OTG) interface, multiple audio interfaces like I2S upholding up to 16 channels per interface, and a broad frequency range of up to 384 kHz.
 

A Dev Kit to Help AI Beginners and Experts Alike

To complement the new series of SoCs, Synaptics also launched the Astra Machina Foundation Series development kit.

 

Astra Machina Foundation Series development kit

Astra Machina Foundation Series development kit. Image used courtesy of Synaptics
 

The kit features a modular design that allows designers to swap compute modules on a common I/O board. This modular nature accommodates various configurations, with daughter cards providing integrated Wi-Fi and Bluetooth connectivity, flexible input/output options, and debugging capabilities. The kit has an open Yocto Linux distribution, including the Synaptics AI toolkit, to facilitate all stages of AI integration.