Seeing the World in 3D: How eInfochips Engineered a Six-Camera Sensor Head on the Qualcomm Dragonwing™ QCS8550 Processor

Customer
A global survey and measurement technology company

Solution Partner
eInfochips Inc., an Arrow company

Application
Multi-camera 3D scanning sensor head for high-resolution 3D data capture and digital twin creation

Industry
Industrial, smart cities, architecture, engineering and construction (AEC), environmental surveillance & emergency response, product monitoring & surveillance

Qualcomm Platform
Dragonwing QCS8550

Featured Offering
Aikri QCS8550: Aikri-85X-50LS-8
Aikri QCS8550: Aikri-85X-50AD-16

Overview

A single handheld device that can walk into a room and hand back a precise, colorized 3D model of everything in it sounds simple. Building the sensor that makes it possible is not. A leading name in survey and measurement technology set out to do exactly that: create a compact “sensor head-pose” capable of capturing dimensionally accurate 3D imagery in real time, using not one camera, but six, a mix of SLAM, RGB, and time-of-flight (ToF) modules working in concert.

To bring that vision to life, the customer turned to eInfochips, an Arrow company and Qualcomm Technologies, Inc., to design the main processing board and a compatible carrier board built around the Aikri QCS8550 System on Module (SoM) powered by Dragonwing QCS8550. The result is a platform engineered to manage six concurrent camera streams, run edge AI processing, and deliver product-level performance in a design ready for volume production.

The Challenge

Most edge AI processors are built to handle a handful of cameras. This design needed three distinct sensor types, each feeding a different part of the 3D reconstruction pipeline, all running at once, all in sync.

• Six-camera fan-in: the sensor head needed simultaneous input from SLAM, RGB, and ToF camera modules, each contributing a different layer of spatial and visual data to build a single, coherent 3D image.
• Beyond native camera ports: the Dragonwing QCS8550 onboard camera serial interfaces couldn't natively terminate all six camera streams at once, so eInfochips had to find a way to bridge the gap without compromising image quality or timing.
• Per-camera ISP pipeline assignment: each of the six streams needed its own image signal processing (ISP) pipeline, correctly mapped and synchronized for concurrent video management, rather than the SoC's default single- or dual-camera pipeline behavior.
• A compact, connectivity-rich enclosure: beyond the cameras, the board had to accommodate a touch display, Ethernet, USB Type-C, an IMU, a barometer and microphone, and a 3.5 mm audio jack; all routed through flex and rigid-flex PCBs into one handheld form factor.
• Production discipline from day one: the design needed to go beyond a proof of concept, with a path through board bring-up, validation, and manufacturing that a real product could be built on.

The Solution

eInfochips answered the six-camera problem with an architecture-first approach — adding an FPGA between a subset of the cameras and the Dragonwing QCS8550, which aggregated the incoming camera streams over the virtual channels of the MIPI/CSI interface. That single decision turned a hardware limitation into a solvable systems problem, and it required restructuring the SoC's internal camera and ISP pipeline, work that eInfochips carried out together with Qualcomm Technologies' engineering team.

From there, the rest of the design came together around the Aikri 8550 SoM (12 GB LPDDR5 / 512 GB–1 TB UFS variant) and a purpose-built carrier board:

• Native support for a 6-camera architecture: the Dragonwing QCS8550 camera interface — 6x 4-lane and 2x 2-lane MIPI-CSI at up to 2.5 Gbps per lane, backed by 3 image front ends (IFEs) plus 2 IFE-Lite paths and always-on capture — provided eInfochips with the interface bandwidth and headroom needed to support the streams aggregated by the FPGA.
• Cognitive ISP pipeline management: the Qualcomm Spectra™ ISP, with triple 18-bit cognitive image processing, were built to let each of the six feeds be tuned and routed independently rather than forcing all cameras through one generic pipeline.
• Purpose-matched camera modules: a Sony IMX900-based SLAM camera (up to 3856×2180 @ 72 fps) for spatial tracking, a Sony IMX678-based RGB camera (up to 2064×1552 @ 60 fps, 8 MP) for high-resolution color imaging, and a ToF module built on the Sony IMX570 (up to 640×480 @ 56 fps) for depth with Qualcomm® Hexagon™ NPU — each optimized for its role in the point cloud.
• Edge AI for real-time reconstruction: the Hexagon Tensor Processor with Hexagon Vector eXtensions (HVX) and Hexagon Matrix eXtensions (HMX), designed to deliver up to 48 TOPS of INT8 performance, processes the sensor fusion and 3D reconstruction workload directly on-device.
• Secure by design: the Qualcomm® Secure Processing Unit provided eInfochips with the interface bandwidth and headroom needed to support the streams aggregated by the FPGA.
• A display solution built for the form factor: rather than defaulting to the more common MIPI DSI display path, eInfochips engineered support for a compact SPI-based display interfaced to the Dragonwing QCS8550 — a harder integration, but one that better suited the device's small handheld footprint.
• High-speed connectivity: a 2.5 Gbps Ethernet link, enabled through an Intel PHY and implemented with Intel's support, plus USB Type-C, provided the bandwidth needed to help offload large 3D capture files.
• A production-ready compute core: the 41.5 mm × 41.5 mm Aikri 8550 LGA SoM gave the program a pre-validated, already-manufactured foundation, so the team could focus its engineering effort on the carrier board, the camera architecture, and the flex/rigid-flex interconnects rather than re-proving the core compute platform.

eInfochips' engagement spanned the full product journey:

• System architecture definition, component selection, and platform design
• Carrier board development leveraging the Aikri 8550 SoM
• PCB fabrication and assembly support
• Board bring-up and functional testing
• Electronics Design Validation Testing (EDVT) and thermal testing
• Driver development and bring-up
• Multi-camera and ISP pipeline management across all six streams

Business Outcome

• Faster time-to-market, lower development cost: building on the production-ready Aikri 8550 SoM and a reusable design framework let the customer avoid re-engineering the compute core, helping cut risk and cost out of the development schedule.
• A working solution to a hard engineering problem: The FPGA-based virtual-channel aggregation architecture helped enable the platform to support six concurrent camera streams, including SLAM, RGB, and ToF data. This approach provided additional interface flexibility beyond the native camera-port configuration and supported implementation of the system architecture used in the demonstration.
• Product-grade security from the start: end-to-end secure boot, built on the Qualcomm Secure Processing Unit, provided a hardware-based foundation for the product's security architecture ahead of certification and launch.
• A display that fits the product, not the other way around: SPI-based display support meant the team didn't have to compromise the device's compact form factor to accommodate a more conventional MIPI DSI panel.
• Enterprise-ready data throughput: 2.5 Gbps Ethernet connectivity, enabled with Intel's support, is structured to give the device the bandwidth to move large 3D capture datasets without the transfer becoming the bottleneck.
• A stable, long-term platform: Qualcomm Product Longevity Program, with support through April 2033, gives the customer a durable compute foundation to build on well past initial launch.
• On track for market: the platform is in active development, with commercial launch targeted for December 2026.