Where MIPI Stops

In the volume applications SUGARS AI designs for — mobile, automotive, drones, robotics, edge AI devices, industrial vision — the sensor output is MIPI CSI-2. Other sensor interfaces exist, at the high end of machine vision and in low-cost parallel-output sensors, but they are the exception, not the path a system integrator designs to. That is a convenience right up to the point where the sensor is not inside a phone.

MIPI has two physical layers, and which one a system uses is decided by the application, not by the system designer. C-PHY carries a symbol across three wires at once, in the differential relationships among them — dense and efficient, and used predominantly in mobile: phone cameras and mobile OLED. D-PHY uses one differential pair per lane for high-speed data, with a low-power signaling mode on the same wires for control, and it is the physical layer most non-mobile MIPI systems use — automotive, drones, robots, industrial vision, and embedded display panels.

The two do not degrade the same way. A C-PHY receiver has to resolve three relationships that attenuate together; a D-PHY receiver has one difference to resolve per pair. Neither specification was written with a channel budget that survives the distances outside a phone, and beyond that budget it guarantees nothing. A receiver front end designed for three simultaneous relationships is not the same circuit as one designed for a single differential pair, and the distance the system has to cover decides how hard that problem is.

MIPI is not the problem. It is the right interface for what it was designed to do, and nothing has replaced it. The constraint is where the specification stops — between the channel the specification assumes and the distance today's systems need the signal to cross. That gap is where SUGARS AI's product line starts.