The parts with no easy replacement

A bill of materials can look healthy right up until the one awkward part disappears. The useful question is not whether every component has a second source. It is which components would force engineering work if you could no longer buy them.

Do that review before purchasing is trying to fill a production order. The output can be short: a list of parts with no checked replacement, what changing each one would disturb, and what you have decided to do about it.

Look for substitution cost, not component price

A resistor may have hundreds of acceptable substitutes. A display costing only a few dollars more can tie together the enclosure, connector, software and user interface. A power-management IC can determine rails and startup sequencing. Flash memory can expose assumptions in boot, update and erase code. A connector with integrated magnetics can be electrically standard on the cable side and unique on the PCB side.

Those are the parts worth discussing. The test is simple: if this exact part vanished tomorrow, what else would we have to change before we could build the product again?

Displays are a good example because the dependency can be hidden inside something you buy as a module. Newhaven published a 2025 change notice for one LCD after the controller IC inside it was discontinued. The replacement kept the same glass and backlight, but changed the controller, instruction set, initialisation sequence, flexible-circuit dimensions and some pin functions. Newhaven recommended that customers validate samples of the new revision.

That is not a reason to avoid display modules. It is a reason to understand what the supplier is free to change and how much of your design follows it. If the display is mechanically central to the product, ask about revision control and product-change notices while you still have options.

Do not call something an alternative until you have checked it

An alternative part number in a spreadsheet is only a lead. An approved alternative is one the engineer is prepared for manufacturing to fit.

For memory, that may mean exercising factory programming, boot, field updates, settings writes and recovery rather than checking only that the board starts. For a display it may include dimensions, connector location, initialisation and the visible result. For a regulator or PMIC it can include startup, transient load, heat and shutdown behaviour. For a connector it may include the PCB footprint, enclosure opening and the electrical function built into the part.

Where a second candidate is genuinely useful, buy a few while the original part is still available. Qualifying it during development is usually easier than doing the same work with a customer order waiting.

Do not turn this into a rule that every line in the BOM needs two sources. Some parts define the product and are not economically interchangeable. Accepting that dependency can be a perfectly good decision. The important thing is to make it deliberately.

Stock can be cheaper than redesign

Suppose an awkward £3 component is used at 100 units a month and a credible replacement would take several weeks of engineering and another board build. Six months of stock is £1,800 before freight and carrying cost. That may be cheap insurance.

The same calculation applied indiscriminately to every component would be nonsense. Hold stock selectively where the substitution cost is high, demand is reasonably understood and the design is stable enough that the inventory is unlikely to be stranded.

Also check what a lifecycle promise actually means. NXP, for example, measures availability under its Product Longevity programme from the product's launch date, not from the date your company starts using it. The programme can also permit migrations to form-fit-function compatible products. A headline such as '15-year longevity' therefore needs to be checked against the exact part and the time remaining.

A catalogue is not the whole supply relationship

When a part matters to the product, ask the manufacturer or authorised channel what commercial supply arrangements are actually available. Depending on the supplier, that might mean scheduled deliveries, an order covering future releases, allocation against a forecast, or simply a named contact who can tell you more than a public stock page.

Raspberry Pi made this distinction very visibly during the 2022 shortages. Retail stock was extremely constrained, while the company said it was prioritising commercial and industrial customers and asked volume users who were struggling to contact it directly. That does not mean a manufacturer can always find parts for you. It does mean that repeatedly refreshing a reseller page is not always the same thing as managing commercial supply.

For genuinely difficult parts, purchasing and engineering should know each other. Purchasing needs to know which substitutions require approval. Engineering needs to know which supply assumptions are holding the design together.

Do one pass over the real BOM

Take the released bill of materials and mark only the components where loss of the exact part would cause meaningful work. For each one, write down four things:

• What else changes if this part changes: PCB, firmware, enclosure, calibration, certification or test?
• Is there a replacement we have actually checked, rather than one that merely looks plausible?
• Would holding stock be cheaper and safer than qualifying another part now?
• What supply or lifecycle information do we have beyond today's distributor stock count?

The result does not need a risk score or a new management system. It needs decisions. Qualify the second flash device. Ask the display supplier how revisions are controlled. Put six months of the awkward sensor on the shelf. Accept that the processor module is a strategic dependency and manage its supply accordingly.

The aim is not a perfectly dual-sourced BOM. It is to avoid discovering, during a production order, that a £4 component quietly owns three months of engineering work.