Almost nobody sets out to redesign working electronics. The board ships, the product sells, and engineering time competes against something more visible. So the design gets patched: a jumper wire here, a firmware workaround there, a substitute part when the original goes short, a heat sink glued on to stop thermal shutdowns. Each patch is individually cheap and individually justified. Collectively they turn a two-week job into a two-quarter one, and the trigger to finally do it arrives as a crisis rather than a plan.
The useful skill is spotting the signals early enough to redesign on your own schedule.
Seven signals a redesign is overdue
1. Components are going end-of-life faster than you can substitute
One obsolete passive is a purchasing problem. A pattern is a design problem. When you have replaced three or four parts in eighteen months, when the microcontroller family has a "not recommended for new designs" flag, or when your buyer is quoting brokers instead of authorized distributors, the board is aging out from underneath you. Broker parts also bring counterfeit risk into a product you are legally responsible for. The single-part version of this is handled in what to do when a component goes end-of-life; the pattern version means the whole architecture is on old silicon.
2. Every change fails certification again
If a small change to the board forces a full re-test rather than a modest delta assessment, the design has no margin. A radio that passed emissions by half a decibel will fail the moment anything moves. Products that sit right on a limit line get re-tested every spin, and each cycle costs weeks and tens of thousands of dollars — a recurring tax that a redesign with real margin eliminates. The cost of the cycle is laid out in FCC certification for electronic products.
3. Yield is drifting down at the contract manufacturer
Watch first-pass yield over time, not in isolation. A board that tested at 97 percent two years ago and tests at 88 percent now, on the same process with the same operators, is telling you something: a marginal timing path, a part creeping toward the edge of its tolerance band, an assembly step that is fundamentally hard. Each failed unit costs the board, the labor, the rework, and often a second test cycle. At meaningful volume a nine-point yield loss can exceed the cost of a full redesign in a year.
4. Thermal problems or field failures cluster
Returns are data. Failures clustered in hot climates, in summer months, or on units mounted in enclosed spaces point to a thermal design that was marginal from the start. The same logic applies to failures clustered around a connector (mechanical stress), around power-up (inrush or sequencing), or after a specific number of operating hours (capacitor wear-out, relay contacts, flash endurance). Bolt-on remedies rarely fix a design with no thermal headroom — the structural options are in thermal management in electronic products.
5. The bill of materials costs more than the market allows
Prices move. A design that was competitive at launch may be carrying an expensive microcontroller you have long since outgrown, a display that has a cheaper modern equivalent, three regulators where one modern part would do, and a layer count nobody has revisited. When your gross margin has quietly eroded from 55 percent to 38 percent through part price increases and competitor pricing, a cost-focused redesign is a revenue project, not an engineering one. Where the money actually hides is mapped in where the money hides in an electronic product's cost.
6. The microcontroller is out of room
A firmware team that has started removing features to fit new ones, disabling debug symbols to make a build link, or writing progressively stranger code to save RAM has hit a hard ceiling. Flash and RAM exhaustion is one of the clearest redesign signals because there is no workaround that does not cost you reliability — and because the fix is often cheap, since a pin-compatible part with four times the memory frequently exists for a small premium. The sizing question is worth revisiting properly, as in how much flash and RAM your microcontroller really needs. The same signal applies to exhausted GPIO, a saturated bus, or an ADC with no channels left.
7. The roadmap needs something the architecture cannot absorb
Sales wants a cellular variant. Customers want over-the-air updates on a device with no spare flash bank. A new market needs a different voltage input or an isolated interface. When the next two years of the product roadmap all require the same structural change, doing it once deliberately beats three consecutive bolt-ons. This is usually the moment the conversation shifts from a board respin to building version 2 of the product.
Redesign or patch: a decision framework
Not every signal justifies a respin. Work through these five questions in order and the answer usually becomes obvious.
| Question | Points toward a patch | Points toward a redesign |
|---|---|---|
| Is the root cause structural? | One part, one net, one firmware bug | Architecture, thermal envelope, memory ceiling |
| How many patches already exist? | Zero or one | Three or more, or any rework step in production |
| What is the recurring cost of not fixing it? | Negligible per unit | Rework labor, yield loss, warranty, or margin erosion |
| How long will this product ship? | Under a year; replacement already funded | Three or more years remaining |
| Does the fix touch certification or safety? | No re-test triggered | Re-test needed anyway — then fix everything at once |
Two numbers turn this into a business case. The first is the fully loaded cost of the redesign: engineering hours, new prototypes, tooling changes, re-certification, firmware port, documentation, and the inventory of old boards you will write off. For a moderate board this typically runs $30,000 to $120,000 all in, with certification often the largest single line. The second is the annual cost of the status quo: rework labor times volume, plus scrap, plus warranty claims, plus the margin you are losing to part price increases, plus the engineering hours consumed by patches and firefighting. When the second number pays back the first inside eighteen months, the decision is made.
The hidden cost of patching: nobody understands the board anymore
There is a cost that never appears in either number. Every jumper, every do-not-populate, every firmware workaround for a hardware quirk widens the gap between the design files and the physical product. Eventually the schematic no longer describes what is on the line, the original engineer has left, and any change becomes archaeology. If your assembly instructions include manual rework steps, or if the answer to "why is that resistor there" is "nobody remembers," the redesign is partly a documentation recovery project — and a disciplined engineering change order process is what stops the new design from ending up in the same state.
Scoping it so it does not become a new product
The biggest risk in a redesign is scope. A project that starts as "replace the obsolete microcontroller" acquires a nicer display, a new connector, and a feature request, and turns into an eighteen-month program. Contain it:
- Write down what must not change — mounting, connector positions, enclosure fit, cable pinouts, user-facing behavior. Everything on that list saves tooling, documentation, and re-testing.
- Decide up front whether it is a drop-in replacement or a new generation. A form-fit-function replacement that reuses the enclosure and passes the same tests is a fundamentally cheaper project.
- Fix the whole list in one spin. If you are re-certifying anyway, take the cost reduction, the thermal margin, the extra memory, and the test points at once. A second spin six months later costs nearly as much as the first.
- Plan the transition. Old and new boards will coexist in the field, so firmware versioning, serial ranges, spares, and service documentation need a plan before the first new unit ships.
- Budget a bridge. A last-time buy of the obsolete part often carries the product through the redesign without a stockout, more cheaply than expediting engineering.
Redesigning something that already sells is a different discipline from new development: tighter constraints, near-zero tolerance for regression, measurable payback. That broader case is made in redesigning an existing product.
Projects House takes on board redesigns for products already in the field — obsolescence replacements, cost reductions, thermal fixes, and architecture refreshes — starting with an honest look at whether a redesign is actually the cheaper path. If your electronics are being held together with patches, tell us what is going wrong through our contact form.