SiMa.ai Hardware Documentation

Use this site for hardware setup, firmware workflows, and low-level interfaces for SiMa.ai MLSoC Modalix products. Software, ML, and pipeline tooling are documented separately.

SiMa.ai Hardware at a Glance

SiMa.ai offers MLSoC Modalix as a chip, on a module, and in complete products. Start from the form factor that you have or plan to use.

Product	What it is	Start here
MLSoC Modalix	The system-on-chip. You place it on a board that you design.	Build with the SoC
Modalix SoM	A module with the SoC, LPDDR5 memory, and power. It plugs into a carrier board.	Build with the SoM
Modalix DevKit	A Modalix SoM on a carrier board, in an enclosure, for evaluation and development.	Evaluate a DevKit
Modalix PCIe Card	A half-height, half-length card that adds Modalix inference to a host machine.	PCIe Mode

To change the system software or add a camera on any of these, see Build System Software and Integrate Cameras.

SoM or Chip-Down

To build your own hardware, choose between a carrier board for the Modalix SoM and a board with the SoC placed directly on it. They need different design work, different collateral, and different bring-up.

	Chip-down design	Carrier board design
What you design	A board with the packaged Modalix SoC on it, including memory, power, and clocks	A carrier board for the Modalix SoM, which already carries the SoC, LPDDR5, and power
Main design work	Power sequencing, clocks and reset, LPDDR5 routing and tuning, high-speed layout, interfaces, thermals	Connector mapping, power delivery, interfaces, thermals
Collateral	Chip-down design guide, SoC datasheet, and design packages, available from SiMa.ai under NDA	Public datasheets and bring-up guide
Software starting point	The Modalix BSP, with board-level adaptation	The Modalix BSP, with device-tree changes for your carrier

After the board boots, both paths use the same system software. See Build System Software.

About the MLSoC Modalix

The MLSoC Modalix is SiMa.ai's second-generation machine-learning system-on-chip. It combines an Arm Cortex-A65 application complex, a high-throughput Machine Learning Accelerator (MLA), an Image Signal Processor (ISP) for camera ingest, and a Computer Vision Unit (CVU) for classical vision workloads — all on a single die.

This integration lets you run multimodal, generative, and vision AI pipelines at the edge without combining separate accelerators, host CPUs, and camera bridges. The Modalix SoM packages MLSoC Modalix for carrier-board and product integration, and every DevKit documented here uses the same silicon.

DevKit Portfolio

SiMa.ai offers MLSoC Modalix-based development kits and PCIe products for edge AI applications. Select a product below, or compare features in the table.

Modalix DevKit

The current development kit for evaluating MLSoC Modalix and building applications on the Modalix SoM.

Modalix PCIe Card

Half-Height/Half-Length (HHHL) card designed for extensible edge compute systems.

Modalix Early Access DevKit

Legacy development kit for existing Early Access customers.

Feature Comparison

Feature	Modalix DevKit	Modalix PCIe Card	Modalix Early Access DevKit
Documentation & OS
Product Brief	view	view	view
Preloaded Operating System	eLxr Linux	eLxr Linux	eLxr Linux
Compute
ARM Cores	8x Arm Cortex-A65 @ 1.4 GHz	8x Arm Cortex-A65 @ 1.4 GHz	8x Arm Cortex-A65 @ 1.4 GHz
ISP (Image Signal Processor)	Arm Mali-C71AE @ 1.2 GHz	Arm Mali-C71AE @ 1.2 GHz	Arm Mali-C71AE @ 1.2 GHz
CVU (Computer Vision Unit)	Synopsys EV74 @ 720 16-bit GOPS	Synopsys EV74 @ 720 16-bit GOPS	Synopsys EV74 @ 720 16-bit GOPS
Memory & Storage
RAM Size	16 GB LPDDR5 (32 GB on earlier kits)	8 GB or 16 GB LPDDR5	64 GB LPDDR5
Storage	16 GB eMMC, 500 GB NVMe	32 GB eMMC	10 GB eMMC
SD Card Slot	✖	✔	✔
Networking
Ethernet	1x 1GbE	1x 10GbE	1x 1GbE (end0), 1x 10GbE (end1), 2x 10GbE SFP+ (end2/3)
Camera Inputs
MIPI CSI	2x 2-lane MIPI CSI	✖	4x 4-lane MIPI CSI
GMSL2 over FAKRA	✖	2x GMSL2 over FAKRA	✖
I/O & Display
GPIO / Headers	40-pin GPIO header	✖	40-pin GPIO header
USB	4 USB 3.2 ports	✖	✖
HDMI	1 HDMI 1.4 port	✖	✖
Graphics Controller	Silicon Motion SM768	✖	✖
Video Codecs
H.264/H.265 Encoder	4kp60	4kp60	4kp60
MJPEG Encoder	4kp30	4kp30	4kp30
H.264/H.265 Decoder	4kp60	4kp60	4kp60
AV1 and MJPEG Decoder	4kp60	4kp60	4kp60
Form Factor
Host-attached PCIe card	✖	✔	✖

Order Your DevKit

Deployment Architectures

MLSoC Modalix supports two common development architectures: standalone mode and PCIe mode. Choose the architecture that matches how your hardware is connected and where your application runs.

Standalone Mode

In standalone mode, a Modalix DevKit or a custom Modalix SoM-based system runs as a self-contained device. Use this mode when the device receives data directly from sensors, cameras, storage, or network interfaces.

Key Use Cases

• Edge AI applications: Run inference without relying on a central server or cloud infrastructure.
• Compact deployments: Reduce the need for additional host hardware.
• Power-constrained systems: Support remote monitoring systems powered by batteries or solar panels.

Advantages

• Self-contained: Runs without a host machine.
• Power efficient: Supports power-sensitive environments.
• Compact: Fits space-constrained deployments.

Typical Data Flow

1. The DevKit receives data directly from network interfaces or sensors.
2. The DevKit loads a NEAT application that defines the on-device inference pipeline.
3. MLSoC Modalix performs inferencing and processes the data locally.
4. Results are sent to other devices or systems via network connections for further action or visualization.

Set up standalone mode

PCIe Mode

In PCIe mode, the Modalix PCIe Card is installed in a host machine. Use this mode when a server, desktop, or workstation handles orchestration, storage, and I/O while the card provides Modalix inference acceleration.

Key Use Cases

• High-performance systems: Use a data-center server or workstation for data processing and storage while the Modalix PCIe Card runs inference.
• Host-managed I/O: Use host storage, networking, or sensor ingest while offloading inference to the card.

Advantages

• Dedicated acceleration: Offload inference from the host CPU to Modalix hardware.
• Expandable capacity: Add inference cards as demand grows.

Typical Data Processing Flow

1. The host machine captures data from sensors, peripherals, storage, or network interfaces.
2. The Modalix PCIe Card loads a NEAT application that packages the inference pipeline and adapted AI model.
3. The host sends data to the Modalix PCIe Card over PCIe.
4. The host processes the results or forwards them to downstream systems.

Set up PCIe mode

Choose Your Path

Evaluate a DevKit

Set up a Modalix DevKit or PCIe card, keep it updated, and start developing.

Build with the SoM

Choose a carrier board for the Modalix SoM, or design and bring up your own.

Build with the SoC

Put the Modalix SoC directly on your own board.

Build System Software

Build the kernel, U-Boot, and device trees, and flash your own image.

Integrate Cameras

Choose a supported MIPI camera, connect it, and select its overlay.

Reference & Downloads

Find product briefs, datasheets, design guides, pin maps, and tech notes.

Already know what you need?

• Build the kernel, U-Boot, and device trees
• Write a device-tree overlay
• Find the source commit for a release
• Flash your own image to a board
• Build an update package
• Compare carrier boards
• Map GPIO, UART, I2C, and SPI pins
• Configure MIPI cameras
• Update or recover a DevKit
• Flash and test boards in production
• Find a datasheet or design document

Get Started

Pick the path that matches your setup. Each guide covers serial-console access, network bring-up, and firmware management for that mode.

Quick Start Guide

Step-by-step interactive walkthrough for first-time DevKit bring-up.

Standalone Mode

Run a Modalix DevKit or Modalix SoM-based system as a self-contained edge device.

PCIe Mode

Install the Modalix PCIe Card into a host machine for offloaded inference.