DRP-AI and Dedicated Vision NPUs: SolidRun's Renesas and Hailo Compute Platforms

TL;DR: SolidRun's Renesas RZ family SOMs and Hailo-15 SOM address two complementary edge AI deployment profiles — integrated DRP-AI acceleration across a scalable ARM Cortex-A55 SOM lineup for IIoT, HMI, and real-time vision applications, and a dedicated vision NPU platform delivering up to 20 TOPS at camera power levels for high-throughput AIoT deployments. OEMs selecting between these platforms should evaluate workload type, integration complexity, and required inference throughput before committing to a carrier board design.

 

The Case for Dedicated AI Acceleration at the Edge

Industrial edge computing has entered a phase where CPU-only inference is no longer adequate for production deployments. Object detection, segmentation, pose estimation, and anomaly detection pipelines impose inference latency and power budgets that general-purpose application cores cannot meet at scale. The industry response has been the proliferation of integrated and discrete neural processing units — each carrying distinct tradeoffs between flexibility, performance ceiling, and integration overhead. For OEMs designing IIoT controllers, smart cameras, HMI panels, and industrial vision nodes, the platform decision is not simply a silicon selection: it determines carrier board architecture, software stack, OS support, and product lifecycle commitment. SolidRun addresses this landscape with two AI-capable platform families that serve different points on the compute and integration spectrum.

The Renesas RZ Family: Scalable ARM SOMs with Integrated DRP-AI

The Renesas RZ product family from SolidRun spans five SOM variants addressing a continuum from cost-optimized Linux compute to vision-capable AI inference. The RZ/G2UL is the entry-level SOM — a single ARM Cortex-A55 core with Gigabit Ethernet, USB, camera input, and serial interfaces, suited to compact and energy-constrained IIoT nodes. Above it sits the RZ/G2LC, offering single or dual Cortex-A55 cores at 1.2 GHz alongside an ARM Mali-G31 graphics engine, CAN-FD, and MIPI video interfaces in the same 47 x 30 mm footprint. The RZ/G2L adds a Cortex-M33 real-time co-processor at 200 MHz, dual CAN-FD, a 3D graphics engine, H.264 video codec, and onboard certified dual-band WiFi 802.11ac with Bluetooth 5.0. All three RZ/G variants are pin-compatible with each other, allowing OEMs to scale or substitute silicon without carrier board redesign.

SolidRun's Renesas platform extends further into vision AI with the RZ/V2L and RZ/V2N SOMs. The RZ/V2L integrates Renesas' DRP-AI accelerator alongside dual Cortex-A55 cores at 1.2 GHz and a Cortex-M33 — delivering on-chip AI inference without heat sink or cooling fan requirements due to its low power envelope. Critically, the RZ/V2L is pin-compatible with the RZ/G2L, enabling OEMs to upgrade an existing carrier design to AI-capable silicon without NRE cost. The RZ/V2N takes the architecture further: quad Cortex-A55 cores running at 1.8 GHz paired with the DRP-AI3 accelerator delivering 15 Sparse TOPS and 4 Dense TOPS, a dual-channel MIPI CSI-2 interface with integrated ISP, PCIe, and USB 3.0. The DRP-AI3 supports face recognition, object detection, surveillance, and quality control inference workloads directly at the edge node, reducing cloud dependency and eliminating round-trip latency.

HummingBoard RZ: From Development to Production

One of the distinguishing characteristics of SolidRun's approach is the availability of HummingBoard RZ SBCs paired directly with each Renesas RZ SOM variant — providing a deployment-ready evaluation and production platform for every silicon tier. The HummingBoard RZ/V2N IIoT SBC delivers dual RS232, dual RS485, dual 1Gb Ethernet RJ45, MIPI-DSI display support, and the DRP-AI3 accelerator in an industrial SBC form factor. The HummingBoard RZ/G2L IIoT SBC targets HMI deployments with CAN-FD, RS232/RS485, Gigabit Ethernet with PoE, and a wide input voltage range. According to published research through the edge inference deployment patterns established by the Industrial Internet Consortium, reducing cloud round-trip latency through on-device inference is among the most impactful architectural improvements available to IIoT system designers. The Civil Infrastructure Platform (CIP) Linux long-term support distribution is available across the RZ/G family, providing a stable, auditable software baseline suited to infrastructure and industrial deployments requiring extended maintenance commitments.

The Hailo-15 SOM: Dedicated Vision NPU Architecture

Where the Renesas RZ family integrates AI acceleration within a general-purpose embedded SoC, the Hailo platform represents a dedicated AI vision processor architecture. The Hailo-15 SOM is built on Hailo's AI vision SoC, available in two variants: the Hailo-15M at 11 TOPS and the Hailo-15H at 20 TOPS. Both deliver best-in-class performance per dollar at camera power levels — the Hailo-15H processes 4K video at 30 FPS using only approximately 1W of power while sustaining up to 3 TOPS for simultaneous AI video analytics. The SOM integrates quad ARM Cortex-A53 cores at 1.3 GHz, dual 2x MIPI-CSI2 4-lane camera interfaces, a 12 MP ISP pipeline with HDR support, H.264/H.265 hardware video encoding, onboard LPDDR4 up to 8 GB, eMMC up to 256 GB, and WiFi 802.11ac with Bluetooth 5.0 — all within the same 47 x 30 mm form factor shared across SolidRun's SOM portfolio.

Software Ecosystem and Framework Support

The Hailo-15 SOM's software ecosystem is a primary differentiator for OEMs deploying production AI vision systems. SolidRun provides a Yocto-based Linux distribution with a Hailo dataflow compiler, a pre-built library of deep learning examples, and native support for Keras, TensorFlow, PyTorch, and ONNX — the standard ML frameworks used in production model development pipelines. The HummingBoard Hailo-15 AIOT SBC extends the SOM into an industrial deployment platform: it includes dual RS232, Gigabit Ethernet with PoE, M.2 B-key slot for LTE modem, dual SIM support, pre-tuned integration with the Sony IMX678 camera module, and a ruggedized extruded aluminum enclosure rated for -40°C to 85°C. As covered in depth by editors at embedded systems design resources covering the embedded systems industry, the combination of ML framework portability and hardware ISP integration is a critical enabler for reducing time-to-deployment in production vision systems. The enclosure measures 120 x 80 x 30 mm with DIN rail mounting support, targeting factory-floor and outdoor edge vision deployments where form factor and thermal management are constrained.

Deployment Profiles: When to Choose Which Platform

The selection between Renesas RZ and Hailo-15 aligns with the fundamental question of whether AI inference is the primary workload or one of several concurrent tasks. The Renesas RZ family is best matched to OEMs building IIoT controllers, HMI systems, industrial gateways, and embedded Linux nodes where general-purpose connectivity — CAN-FD, RS485, Ethernet, camera, display — is the primary design constraint and AI inference is an enhancement capability. The pin-compatible SOM family allows OEMs to begin with an RZ/G2L or RZ/G2UL baseline and upgrade to RZ/V2L or RZ/V2N when AI inference requirements emerge, without carrier board redesign. The Hailo-15 SOM addresses deployments where vision AI throughput is the primary design requirement — smart cameras, surveillance nodes, retail analytics platforms, medical imaging devices, and any application where sustained 4K inference at low power is non-negotiable. Its dedicated NPU architecture and ISP pipeline are purpose-built for this workload class in a way that general-purpose SoCs with integrated AI accelerators are not.

SolidRun's OEM Support Model Across Both Platforms

Both platform families reflect SolidRun's one-stop-shop approach to OEM product development. Full hardware documentation, reference carrier board designs, Yocto BSPs, and Developer Center resources are available for both the Renesas RZ and Hailo-15 lines. SolidRun offers assembly options — configurable memory, storage, connectivity, and temperature grade — allowing OEMs to tailor each SOM to application-specific cost and supply chain requirements. Customization and branding services, including NRE options for custom carrier boards and enclosure design, extend both platforms from evaluation kit to production-ready embedded product. Whether the target deployment is a DIN rail-mounted IIoT gateway built on the RZ/G2L, a quad-core vision inference node running DRP-AI3 on the RZ/V2N, or a compact industrial AIoT camera platform powered by the Hailo-15H, SolidRun provides the engineering support and hardware ecosystem to accelerate the path from prototype to production without unnecessary overhead.