TI’s Video SoC Dilemma

We’ve had the pleasure to work with Texas Instruments’ high-end video processor families (DM8168 and DM8148) over the past few years on a number of different projects. These are highly integrated Systems-on-Chip (SoC) that contain many independent subsystems such as an ARM Cortex CPU, a DSP, video compression/decompression engines, a 3D graphics accelerator, and a large video processing subsystem capable of handling multiple HD video streams simultaneously. These types of devices hold the promise of rapidly bringing to market complex video products in a compact form factor and at low cost.

The TI products aren’t really appropriate for broadcast video applications, but they’re ideal for a wide range of applications in enterprise video, law enforcement, defense, and other markets.

The problem, however, is that TI is struggling with how to support these complex devices. Over the past 12 months, TI has backed off releasing certain promised video features and has changed their support model for these chips several times. It appears that TI has decided to only offer these devices to higher volume customers, and will support lower volume clients only with internal approval. This is not uncommon in the video space. Specialty video processor vendors such as Ambarella, Cavium, and Maxim (the former Mobilygen) have always targeted specific video markets and have minimum engagement policies with their customers. TI had been different, offering development boards to any willing customer through their distributors and providing technical support to most developers. That’s been great for mid-volume products.

TI’s ability to design high-end video chips has clearly outpaced its ability to support their customers who are integrating them into complex products. This is not surprising — the technical reference manual for the DM8148 is over 3000 pages long! The development board contains demonstration applications for basic video operations such as capture, encode, decode, and display. However, developing custom features that take advantage of the many less common video modes can be a huge time investment and can require lots of support.

As everyone knows, SoCs will continue to become more and more highly integrated and complex. The problem of offering complex devices to a large development audience will continue to grow. These platforms have come a long way by leveraging open-source operating systems such as Linux and Android. They have also made strides by conforming to mid-layer standards such as GStreamer and Qt. However, these solutions only support a fraction of the capabilities of the devices, and developers often need to use other features in order to differentiate their products.

TI may be coming late to the realization that their business case doesn’t work to support complex SoCs with small volume sales. Ambarella et al have figured this out long ago. Being a larger company with a bigger customer base, TI has done what they can via their crowd-sourced E2E support website as well as having third-party design firms (such as Cardinal Peak) provide support to lower-volume customers. However, it appears that they are still figuring out how best to respond to these challenges.