原创 Struggles of global SoC design teams

Communications between design teams have never been easier. We have real-time collaboration, video-conferencing, the cloud, virtualisation, mobility, texting, email, VoIP, and source control.

Everybody is now available anywhere at any time of the day.

We have the ability to track what everybody is doing and devise new ways of improving productivity.

Why, then, is it still so difficult to complete System-on-a-Chip designs?

The process of integrating the efforts of globally distributed design teams into a single SoC is still a daunting task. It was identified as a top challenge in a recent survey of Arteris customers, and it is easy to see why. Integration of the work of distributed teams is a major cause of delays in getting to market. Delays represent lost revenue and missed opportunity.

No doubt all of these revolutionary collaboration platforms have helped the industry reduce difficulties in communication between geographically distributed design centres. They have lowered travel barriers, eroded time zone differences, and cut through geographical divides.

Any company can enable its top designers to work anywhere in the world.

Geographically dispersed SoC design teams.

However, bringing all that work into a final design and getting through the top level of performance and functional verification is still causing anxiety.

As many devices surpass 300 million gates and 100s of IP blocks, it is easy to understand why there is complexity involved in bringing it all together.

What is making global management even more difficult is the specialisation that has evolved with each design team. One team will focus on the CPU complex. Another will concentrate on the memory sub-system, and so on for graphics, video, audio, I/O, etc.

As each design centre makes revisions to a specific portion of the chip, it inadvertently affects the entire SoC.

Additionally, the tasks for these specialists have become more complex. The sections of each chip have become full-blown sub-systems. Complexity is on the rise, and each of these sub-systems now has multiple IPs and its own interconnect. The specialists are focused on power, performance, and area optimisations of their own sub-systems. In so doing, they focus on the verification of their portions of the SoC.

Once this is complete, another team stitches all the different sub-systems together in a top-level SoC interconnect fabric. If some sub-systems are not cooperating with the rest of the chip, someone has to go back to the sub-system design teams to determine why and figure out what fix can be made.

SoCs that use a hybrid bus or crossbar architecture within the interconnect fabric are particularly prone to delays in this part of the process. Bridges and shim logic are required at IP connections to make the separate sub-systems work harmoniously with each other. Reassembling these sub-systems often results in timing issues, protocol and address mismatches, and power management bugs.

Many global design teams have resigned themselves to the fact that the integration stage of development is an inevitable part of the process. However, the additional time and complexity that the process adds represents lost revenue and lost opportunity. More importantly, these factors are completely avoidable.

As single platform strategies gain more traction, SoC architects need ways to divide and re-assemble chips by allowing each IP sub-system design team to work independently.

After each team of specialists completes its design on an individual sub-system, this new way of collaborating should re-assemble and connect all of the different sub-systems seamlessly and automatically. This enables the integrated SoC design team to complete top-level verification more quickly and get the complete chip design to market sooner.

With so many means of real-time collaboration at our disposal today, design managers have to decide which ones match the needs of an organisation by asking for each one:

Can it cut cost, time, and complexity? Does it help geographically dispersed teams work together more harmoniously? Can it help organise talent by expertise? Can specific tasks be assigned according to that expertise? If it could do all these things, how much would that be worth to the organisation?

Design and communication tools alone cannot meet the requirements to help make design-team collaboration more efficient. Design teams need to architect SoCs with the requirements for worldwide development in mind, and use interconnect IP that enables distributed SoC development.

Kurt Shuler is vice president of marketing at Arteris and has extensive IP, semiconductor, and software marketing experience in the mobile, consumer, and enterprise segments working for Intel, Texas Instruments, and three start-ups. Prior to his entry into technology, he served in the US Air Force Special Operations Forces.