Semiconductor industry
From silicon to systems: where engineering creates value
A practical view of the semiconductor value chain, its interfaces, and the decisions that connect design to a working product.
Start with the complete product
A semiconductor is often introduced through its transistor count or process node. Those are useful details, but they explain only a fraction of the work required to deliver a dependable product. A chip must be designed, manufactured, tested, packaged, integrated, supported and supplied. A weakness at any interface can erase an advantage elsewhere.
I find it useful to follow two journeys together: the physical journey of the device and the information journey that defines it. Silicon moves through facilities. Requirements, models, design databases, test limits and qualification results move between teams. The two journeys must agree before a customer can rely on the result.
This is why semiconductor engineering rewards systems thinking. A smaller die may improve wafer economics while making power delivery harder. A cheaper package may constrain bandwidth. A feature that looks attractive in a product presentation may add verification effort, firmware complexity and test time. The design task is to resolve these interactions deliberately.
Five connected stages
| Stage | What it contributes | What the next stage needs |
|---|---|---|
| Materials and equipment | Repeatable physical capability | Purity, consistency and qualified processes |
| Design and IP | Function, architecture and implementation | Complete constraints and verified design data |
| Wafer fabrication | Devices and interconnect | Traceable process and electrical results |
| Packaging and test | Usable, screened components | Electrical, thermal and mechanical compatibility |
| System integration | A customer application | Software, documentation and lifecycle support |
The sequence is a map of responsibility, not a strict waterfall. Package planning begins while architecture is still flexible. Test access affects the logic design. Firmware teams need models before silicon arrives. Reliability requirements shape device, process and supplier choices long before qualification testing starts.
Fabrication itself is also iterative. Deposition, patterning, etching and measurement recur across many layers. Inspection and process-control data feed back into manufacturing decisions. Treating the fab as a single box hides the learning needed to make a design repeatably.
Ownership changes; accountability remains
A fabless company owns product definition and design while relying on manufacturing partners. An integrated device manufacturer retains more manufacturing capability. A foundry specialises in manufacturing for customers, while an assembly and test provider supplies another part of the chain. EDA and IP suppliers support multiple stages.
These models distribute investment and expertise differently. They do not remove the need for integration. A fabless business still needs people who can evaluate yield, test coverage, package options and supply commitments. Outsourcing a process does not outsource responsibility for the customer outcome.
The same principle applies inside a design organisation. A team can license a processor core without acquiring a finished software platform. It can purchase an interface controller without resolving board signal integrity. It can obtain a process design kit without possessing production experience on that process. Every purchased capability comes with assumptions that must be checked against the intended application.
Make the interfaces visible
For a new product, I would map the highest-consequence interfaces before optimising individual blocks. Which requirements are still ambiguous? Which supplier deliverables sit on the critical path? Which assumptions about bandwidth, temperature or software response could change the architecture? Which failures can be detected before committing to masks or production inventory?
A compact interface register can record the owner, input, output, acceptance criteria and change notification for each hand-off. This makes a review concrete. Instead of asking whether packaging is on track, ask whether the pin map, power envelope, mechanical limits and test access have been agreed by the teams that depend on them.
The most useful engineering decision is often the one that prevents a later mismatch. A clear requirement avoids an unnecessary feature. A realistic power budget prevents a thermal redesign. A test strategy exposes an inaccessible fault. A well-defined software interface shortens bring-up.
That is the value-chain perspective I bring to semiconductor work: understand the local problem deeply, then check its consequences across the complete product. Precision at the block level and coordination at the system level belong in the same conversation.
Further reading
Sand to System
Explore the broader ideas that connect this article to the semiconductor industry and the practice of engineering.
Explore the fieldnotes