SoCs: when to make versus when to buy

Many internet-connected embedded products today run some flavour of Linux, and Linux requires a reasonably sophisticated System on Chip (SoC). The underlying hardware implementation typically boils down to the venerable CPU-RAM-ROM triad, a motif found under the hood of virtually every computer for decades. This triad of components is technically challenging to implement, yet ironically is one of the least differentiating features.

Product developers that prioritise BOM cost reductions will blindly throw extensive engineering resources at refining and cost-optimising the CPU-RAM-ROM triad. It seems easy to justify throwing resources at a full-custom solution that can shave tens of dollars off a BOM, which would otherwise go to a System on Module (SoM) vendor like Raspberry Pi, Variscite (www.variscite.com), or SolidRun (www.solid-run.com). This is sound logic if you’re a large corporation gearing up for production runs of hundreds of thousands of units.

However, even the most successful startup products typically run only a few thousand units before requiring a major design revision: bold new ideas make bold assumptions that rarely survive contact with end customers. Fresh clarity on market requirements and user feedback often necessitates changes. In a fail-forward-fast environment, cost-optimising a CPU-RAM-ROM triad doesn’t make sense. It would be like assigning top software developers to first hand-optimise code loops, instead of experimenting with new features and user requirements.

When design requirements are bound to change within the span of a few thousand units, spending tens of dollars extra per unit to buy a modularised CPU-RAM-ROM triad from SoM vendors amortises nicely, compared to the costs required to roll one from scratch. Starting with a well-validated processing core also reduces the risk of schedule delays due to unforeseen complications, while freeing up engineering resources to work on features that are truly unique to your product.

A real life example of this philosophy is Formlabs’ choice to use a SoM inside the Form 2 (read the full tear-down here: hsmag.cc/MPWHAP). Despite having ample venture capital and engineering resources, they wisely focused their efforts on building a better printer, instead of a cheaper or better computer. The wisdom to see past the BOM, and to focus on the issues that make or break a product, is arguably one of the essential competencies that any startup needs if it hopes to eventually grow into a billion-dollar business.

More features from HackSpace magazine magazine

Subscribe