SGM313

One chip between the camera and the AI. Nothing else.

SGM313 extends a complete camera link — the MIPI D-PHY data lanes and clock, plus I²C — beyond 2 m, with no protocol conversion at either end. The host reads and writes the sensor's registers directly, as though the sensor were on the same board.

In development — engineering samples January 2027

Introduction

Take a 4K sensor off the board and onto a cable, and the D-PHY link stops working long before the camera is where the system needs it. D-PHY is specified for the interconnect distances inside a phone; beyond that, the specification guarantees nothing. The common answer is encapsulation: the video is placed inside another protocol — automotive SerDes, an Ethernet-based link, USB — carried, and reconstructed at the far end. Each brings a cost at both ends. Vision Link IC does not encapsulate. It carries the MIPI signal itself, on dedicated conductors, and lets the host read and write the sensor's registers directly. Nothing is added to the stream, so nothing has to be taken back out — and what arrives at the far end is the sensor's own output, not a reconstruction of it.

Key Features

What makes SGM313 different

What the IC does

SGM313 extends a complete camera link — the MIPI D-PHY data lanes and clock, plus I²C — beyond 2 m. The D-PHY signaling is not encapsulated in another protocol at either end, and the host reads and writes the sensor's registers over I²C as though the sensor were on the same board.

Example wiring — the connector is USB Type-C. The link is not USB

A standard USB Type-C cable carries exactly what this link needs: six differential pairs, power, and two spare conductors. The four high-speed pairs carry the data lanes, one more pair carries the clock, the two sideband conductors carry I²C, and VBUS carries power. On the host side it wires to the processor's D-PHY port, not to a USB port. Nothing about SGM313 requires USB Type-C — it is one connector, manufactured by the hundreds of millions, that presents the pairs the link needs at a commodity price. The cable is built to the link's specification, not to USB's.

Two cameras on one chip

Many sensors run on two data lanes and one clock lane, which leaves two of the four data lanes idle. SGM313 supports two clock lanes, so one device carries two independent camera links — two data lanes and a clock apiece. One device and one connector carry two cameras, each at 4K 60 fps in RAW12, uncompressed. With a single four-lane camera, the same link carries 4K 120 fps in RAW12.

Uncompressed, and no added latency

SGM313 adds no line or frame buffer. Link latency is cable propagation plus the device's own fixed delay — not a stored frame. In the display direction, the same link carries 4K 60 Hz in RGB888. Neither direction is compressed.

Nothing to protect against, nothing to write

Because nothing is superimposed on anything else, there is nothing to protect against — none of the power-over-coax filter network and coupling capacitors a coax SerDes needs. Because I²C is physically continuous, there are no serializer/deserializer registers to align and no link driver to write; the camera's device driver stack is unchanged. For two cameras, the bill of materials is still one low-power chip and one cable.

Specifications

SGM313 at a glance

Interface MIPI D-PHY — one 4-lane link, or two independent 2-lane links (one clock lane each)

Reach 2 m and beyond, no protocol conversion

Wiring & connector Ordinary cabling. USB Type-C connector; the link is MIPI D-PHY, not USB; cable built to the link's specification

Single camera (4 data + 1 clock) 4K 120 fps RAW12, uncompressed

Dual camera (2 data + 1 clock each) 2 × 4K 60 fps RAW12, uncompressed

Frame sync Supported

Control I²C direct, one address per sensor — no tunneling, no driver development, camera device driver unchanged

Display direction 4K 60 Hz RGB888, uncompressed

Latency Cable propagation plus a fixed device delay — no line or frame buffer

Power consumption Low

Devices per link One, whether the link carries one camera or two

Status In development — engineering samples January 2027

Applications

Where it is used

Edge AI devices

Cameras at a distance from the inference processor

Drones and robots

Including physical AI systems

Open-source hardware platforms

Single-board computers and AI modules that expose a MIPI D-PHY port

Industrial vision & displays

Vision over a cable rather than on a board; displays driven at a distance in RGB888

For further details and datasheets,
contact sugars@sugars.kr

Datasheets, evaluation boards, and engineering samples are available on request.

Contact Us