COM-HPC Academy: COM-HPC Carrier Design Considerations
December 16, 2020
There’s been a fair amount of hype recently surrounding the coming COM-HPC modules. And for a good reason. PICMG is on the verge of releasing this new spec that pushes the performance of computer-on-modules (COM) to heights not previously attained. However, nearly as important as the COM-HPC module is the carrier board that it plugs into.
The carrier card generally offers a base set of features. Then the COM enables the system for the intended application by adding specific functionality. The COM could be viewed as a replaceable component, which can be upgraded to increase performance or add features that simply weren’t available previously. That could be increased compute power or additional I/O. And if the standards are adhered to, that module could come from any supplier and plugged into any compliant carrier board.
According to Chris Engels, a Product Marketing Manager for Avnet Integrated/MSC, “By using this computer and module philosophy, you end up with a very flexible system. You could connect specific interfaces, like sensors, cameras, or human machine interfaces, displays and so on. The result is a system that’s widely usable over a long period of time.”
The carrier board should first consider the mechanical requirements, such as the housing, packaging, cabling inside and outside the system, the power distribution from external sources, and environmental requirements such as shock and vibration and thermals. An early design decision is whether you plan to design for a COM-HPC server interface or client interface. Then, you must determine the number of modules needed. And finally, a choice must be made regarding which size COM-HPC module to support, as multiple form factors are available. Note that the client and server interfaces require different gaps between the connectors. And precision is of the utmost important as we’re now dealing with very high speed signals.
Looking at the two possible choices for COM-HPC, the client interface provides four graphics interfaces, which can come out of a system-on-chip that’s located on the module. There’s also sufficient real estate, and PCI Express and Ethernet for the performance class of next-generation platforms. However, for applications that require even more PCI Express and Ethernet line, the COM-HPC server interface is likely the right choice. Here, you don't have the graphics interfaces, but you do have more footprint for other I/O. Determining the interface at the early stages of the design process is highly recommended.
The specification defines some signal-integrity recommendations, which help you to keep within the boundaries of the maximum trace lengths and the loss budgets. A host of simulations have been performed to prove this out. Simulations have also been done on various PCB materials.
The ubiquitous USB is another potential method of high-speed signaling. The standard looks at how to implement the latest USB spec, and is also backward compatible with older forms of USB, as well as Thunderbolt or Display Mode.
Notes Engels, “There's also some software work involved, and it really depends on the support devices on the selected COM-HPC module. You’ll need to work with the module supplier to find the right architecture and find the right technical solutions.”
When it comes to cooling, the specification stipulates a heat spreader, which works fine when you’re not driving up to the maximum power levels. In those instances, you may want to consider heat sinks or fans. However, it's implemented, careful planning is required so that you have enough space for air to flow. It’s always a good idea to do thermal studies, with thermal simulation and testing.
To see and hear lots more detail on each of these topics as the relate to the carried-board design, check out the video featuring Avnet Integrated/MSC’s Chris Engels. Also look into starter kits and reference designs offered by a host of vendors.