Following on the heels of a recent post highlighting trade-offs in HDI technology and layer count of a single-board versus a som solution and how that drives bare_board costs, we wanted to take the next step and examine the Total Cost of Ownership (TCO) of a single board development approach vs. incorporating a SOM in your design.
The analysis below looks at both the recurring costs and the engineering investment for the two approaches. The SOM used in this comparison is part of Critical Link’s MitySOM-5CSx family, which is based on the altera cyclone_v_soc, featuring single or dual hard-core arm cortex-a9 (single core for this exercise) tightly integrated with fpga fabric.
Your parts BOM - this represents the total BOM cost for the solution.
Includes all components for the CPU/FPGA infrastructure (assumed in this case to be in the $60 range) in addition to all the components that fill out your total solution. The additional parts could be ADCs, DACs, PHYs, connectors, power supply components, passives, etc. These additional components are assumed to be in the $90 range for this exercise.
The same as the single board solution, minus the components that would be provided as part of the SOM itself.
Bare Board
Approximately 5"x 6" bare board, 12 layers, HDI technology including blind and buried vias, microvias, tight trace and space.
Approximately 5"x 6"bare board, 6 layers, non HDI technology, no blind and buried vias, no microvias, and relaxed trace and space
Assembly - Labor cost for a contract manufacturer to assemble the representative board.
Slightly cheaper than the single board solution since there are fewer components to place
X-Ray - X-Ray inspection of BGA components post assembly
Cost not incurred in this approach. This assumes all BGA components are in the CPU/FPGA subsystem and thus on the SOM.
Test - Functional, flying probe, or other post assembly test method
Slightly less expensive since there are fewer components to cover with testing.
As you can see, the total recurring costs for the single board solution in this mythical scenario is $62 per unit less than the cost for a SOM solution. Even with a far lower cost for the bare board; lower costs for assembly and testing; and the elimination of the need to x-ray, the single board solution looks like the winner.
Let's now extend our scenario to include the cost to engineer each solution. For the purposes of our imaginary scenario let's assume that the cost to engineer the portion of the design already found on the SOM is $250,000 and those portions of the design not covered by the SOM (predominantly I/O, data acquisition, etc.) cost $50,000 to develop.
One may consider this to be an inflated number. In our experience, however, it's pretty aggressive. To illustrate this point consider that this cost includes the total cost of the following (including materials):
Using the analysis shown here, at an annual volume of 1,000 units the TCO of each SOM-based unit produced is more than $100 less than that of a single board, chip-down unit. The break-even point is around 3,000 units per year, assuming a fairly typical target ROI period of 18 months, though most designs don’t get to that kind of volume until a year or more into production.
As you can see, with a SOM based approach, a very significant engineering investment has already been made swinging the cost trade balance in favor of the SOM. Does this analysis hold true for all projects? Of course not. However, for a large variety of industrial and medical type applications it does.
Let's consider additional "behind the scenes" benefits:
If you're being pushed to add complexity to your products despite tight timelines, it may be time to consider the total cost of ownership and value-added benefits of a SOM approach, and perhaps relieve a little bit of resource stress along the way!