⚡ Spark Academy53 lessons

Nothing Is Ideal

Every part you buy is a cloud of possibilities, not a number. Design for the cloud or be surprised by production.

lesson 1 of 2 in this unit

Builds on: 1.3 Resistors in Practice2.2 Voltage Dividers1.4 Power & Energy

The lies simulators tell

Every lab in this course — and every SPICE simulation ever run — used ideal parts: resistors that are exactly their value, wires with zero resistance, capacitors that are only capacitors. Real parts are messier in ways that eventually matter:

  • Tolerance: your ±5% gold-band resistor (1.3) is a promise, not a value: a 10 kΩ part is somewhere between 9.5 k and 10.5 k. Stack several in a circuit and the uncertainties combine.
  • Temperature: resistance drifts with heat (~100 ppm/°C for common parts); a ceramic capacitor can lose half its capacitance between winter shed and summer car.
  • Parasitics: every wire is a small resistor and a small inductor; every capacitor hides series resistance (ESR) and inductance; every adjacent pair of traces is a tiny capacitor. Below ~1 MHz you can mostly ignore this; your PWM's crisp edges (spectrum lesson!) cannot.
  • Ratings and datasheets: the manufacturer's datasheet is the part's contract. The master habit: design to the recommended operating table, treat absolute maximum as the cliff edge you never picnic on.

Designing for the cloud

One prototype working proves that one point of the tolerance cloud works. Production builds thousands of points. Two professional defences:

worst-case: check the corners · Monte Carlo: roll the dice thousands of timesthe lab below builds 800 “identical” dividers and lets you meet the cloud in person

The design conclusions are wonderfully practical: make behaviour depend on ratios of like parts where possible (ratios track better than absolutes — the trick op-amp circuits exploit), buy precision only for the parts that set accuracy, and leave margin everywhere else. A ±3% output spec built from ±10% parts is not frugality; it’s a scrap bin.

One rule worth its own box: decoupling

Every chip's supply pin gets a 100 nF capacitor, millimetres away, to the ground plane. Fast digital edges demand gulps of current that the distant supply (through its parasitic wire inductance — Lesson 2.4!) cannot deliver in time; the local capacitor is the chip's personal reservoir. The 555 datasheet asks for it, the Pico has a forest of them, and forgetting it is the classic source of “works on the bench, glitches in the field.”

⚡ Lab — Meet the Cloud

A production line building 800 “identical” 4.5 V dividers from real, toleranced resistors.

  • ±1% parts: a tight spike, everything in spec. Ship it.
  • ±5%: the bell curve spreads to the spec fences. Count the red tails.
  • ±10%: read the scrap counter. Precision parts suddenly look cheap.
The engineering lesson
one prototype working proves nothing — designs must work across the whole tolerance cloud
Try this
At ±10%, what fraction of production would you scrap? Now you know why precision parts cost more.

Check your understanding

Q1. A ±5% 10 kΩ resistor guarantees…

Q2. Why does one working prototype not prove a design?

Q3. A 100 nF decoupling capacitor next to every chip exists to…

Q4. 'Absolute maximum ratings' in a datasheet mean…