Reading a Datasheet Without Getting Fooled
Typical is a marketing number. Minimum and maximum are engineering. The conditions column is the law.
🎯 By the end you'll understand
The difference between absolute maximum ratings and recommended operating conditions, and why a part can be simultaneously "within its ratings" and outside every guarantee on the page. What typ, min, and max actually promise, which one to design to for each kind of parameter, and how to spot a number that is characterised rather than tested. Why the conditions column beside a value is the value's other half, and why the graphs at the back are often more useful than the tables at the front. How to read the specific pages of the ESP32-C3-MINI-1, SHT40, and LIS2DH12 datasheets that Beacon depends on, why the C3 module was chosen over an ESP32-S3, and how a single misread condition turns a three-year battery estimate into a four-month one without anything failing.
KEY INSIGHTA datasheet is a contract, and most of it is the conditions under which the promises hold
Nobody lied. The number was on the page. It was on the page beside a temperature, a supply voltage, and a mode of operation, and the design satisfied none of the three.
Here is the failure this lesson exists to prevent, and it is not a dramatic one. A team designs a battery-powered sensor. They open the microcontroller's datasheet, find a sleep-current row, take the number, add up a budget, and promise the customer a year of battery life. The product ships. Six months later units start going quiet in the field, and the returns show the batteries are flat. Nothing broke. The number they took was a typical at 25 °C in the deepest sleep mode with RTC memory powered down, and their product runs in a warm ceiling void with RTC memory retained, where the same current is several times higher. There is no defect to find, no rework to do, no supplier to blame. The specification was read without its conditions, and the entire product was designed on top of that reading. Fixing it means a new battery, a new enclosure, or a new product.
Every number in an electrical characteristics table has three parts: the value, the statistic (min, typ, or max), and the conditions. Take any one of them away and the number is meaningless. Typ is usually the mean of a characterisation sample at 25 °C and nominal supply, and it is frequently not tested in production at all — it is a description of a batch, not a promise about your unit. Min and max are the promise, and they are the only numbers you can put in a design margin. And absolute maximum ratings are not a second set of operating limits: they are the stress levels at which the manufacturer stops guaranteeing that the part survives, and nothing at all is guaranteed to work there.
CORE CONCEPTThe anatomy of a datasheet, and the order to read it in
Datasheets are laid out in the order that is convenient for marketing, not the order that is useful for design. Learning to jump is most of the skill.
| Section | What it actually is | Read it… |
|---|---|---|
| Front page: features, block diagram | Marketing. Selected best-case numbers with no conditions. Useful for deciding whether to keep reading. | First, sceptically |
| Absolute maximum ratings | Stress limits. Beyond these the part may be permanently damaged; nothing is guaranteed to work at them. | Second — it bounds your protection design |
| Recommended operating conditions | The envelope inside which every other guarantee is valid. Your design must live entirely inside it. | Third — this is the real specification |
| Electrical characteristics | The guarantees, each with min/typ/max and a conditions column. | Fourth, one row at a time |
| Timing diagrams and tables | Setup, hold, rise, fall, and bus timings. Where protocol bugs live. | When you write the driver |
| Typical performance curves | Functions, not points. Often the only place the temperature and voltage dependence appears. | Whenever a table value looks decisive |
| Application information | The vendor's own reference design, layout advice, and worked examples. Usually excellent and usually skipped. | Before you draw the schematic |
| Package, ordering information | Which part number is which grade, temperature range, tape and reel. | Before you order — grades differ |
| Revision history | What changed and when. A parameter that got worse between revisions is a story. | Every time, and compare to your copy |
| Errata (separate document) | Known silicon bugs and their workarounds, indexed by chip revision. | Before you debug anything |
Two habits follow from that table. First, read the recommended operating conditions before the features, because they tell you whether this part is even a candidate. Second, find the errata document separately. Errata are rarely inside the datasheet; they are a different PDF, sometimes a different page of the vendor's site, and they are indexed by silicon revision — which means you also need to know which revision is printed on the chip in front of you. Espressif publishes errata for the ESP32-C3 family; ST, TI, and Nordic all do the same for their parts. A day spent chasing a peripheral that "should work" is a day you did not have to spend.
When a specification changes between datasheet revisions, it is almost never because the silicon improved. It is because characterisation found something. A quiescent current that moved from 5 µA to 8 µA, a dropout voltage that grew, a timing that gained a footnote — each of those is the manufacturer telling you what they discovered after release. If your design's margin depends on a parameter, check whether that parameter has moved across revisions, and design to the worst of them. It also tells you something about how mature the part is, which matters when you are committing 25 units and then 25,000.
DEFINITIONAbsolute maximum versus recommended operating: the most misread page
Almost every datasheet carries a version of this boilerplate under the absolute maximum ratings table, and it is worth reading slowly at least once in your life:
Stresses beyond those listed may cause permanent damage to the device. These are stress ratings only; functional operation of the device at these or any other conditions beyond those indicated under recommended operating conditions is not implied. Exposure to absolute-maximum-rated conditions for extended periods may affect device reliability.
Three separate claims are packed into that paragraph, and each one is routinely misread.
- "May cause permanent damage." Absolute maximum is the boundary of survival, not of operation. A pin rated to an absolute maximum of 3.6 V will probably survive 3.6 V. It is not guaranteed to do anything correct at 3.6 V.
- "Functional operation … is not implied." Every number in the electrical characteristics table is void outside the recommended operating conditions. If your supply can reach 3.7 V and the recommended range tops out at 3.6 V, then above 3.6 V the ADC's accuracy, the oscillator's frequency, the I/O levels, and the current consumption are all unspecified. Not "slightly worse" — unspecified.
- "For extended periods may affect reliability." Damage is often cumulative and latent. A pin that saw 5 V for a second during bring-up may work for months and then fail, and the failure will look like something else entirely.
There is a fourth rule that is stated less often and matters just as much: absolute maximum ratings generally may not be applied simultaneously unless the datasheet says otherwise. A part rated to 125 °C junction and to some maximum supply voltage is usually not rated to both at once. Where a datasheet gives a safe operating area or a derating curve, that curve is the simultaneous rating, and it supersedes the individual numbers.
This sentence has ended more products than any other in hardware engineering. Running a 3.3 V part from a 3.6 V rail because "abs max is 3.9 V" means every guarantee on the page has been surrendered in exchange for nothing. The correct use of absolute maximum ratings is defensive: they tell you what your protection circuitry has to prevent. The USB VBUS TVS diode in Beacon's canon exists to keep transient events below the absolute maximum of everything downstream. The recommended operating conditions are what your design must satisfy. Two different tables, two completely different jobs.
HOW IT WORKSTyp, min, max, and the column beside them
What each statistic promises
Typ is the centre of a distribution measured on a characterisation lot at 25 °C and nominal supply. It is not usually production-tested. It is the right number for estimating how a batch of a thousand units will behave on average, and the wrong number for promising anything about the unit in your hand or about a unit at 45 °C. When a datasheet gives only a typical for a parameter — and this is extremely common for supply currents, leakages, and RF power — that is the manufacturer declining to guarantee it, and you should design as if it could be two to three times worse.
Min and max are the guarantee, usually across the full recommended operating range including temperature. These are the numbers a production tester screens against, so a part that violates them is supposed to be scrapped. But read the footnotes: many datasheets mark some limits "guaranteed by design" or "guaranteed by characterisation, not production tested", which means the limit is a statistical statement about the process rather than a screen on your unit. That is a weaker promise, and where it appears on a parameter your design depends on, it is worth adding margin.
Which one to design to
| What you are budgeting | Use | Why |
|---|---|---|
| Current consumption / battery life | max, or typ × 2–3 if only typ is given | You must survive the worst unit in the worst corner of the field |
| Output drive current, source or sink | min | The weakest part must still drive the load |
| Input high / low thresholds | Worst end of each | Your driver must exceed VIH(min) and get below VIL(max) |
| Regulator output voltage | Both min and max | Downstream parts have their own recommended range at both ends |
| Timing setup / hold | The worst corner across temperature and voltage | Bus errors are the least reproducible bugs you will ever chase |
| Sensor accuracy against a requirement | max error over the range you use, not typ at 25 °C | Beacon's R3 is a requirement, and requirements are met at the worst case |
| Thermal / package rating | max, derated for your ambient | Lesson 1.2's derating curve is the real limit |
The conditions column is the other half of the number
Beside every value sits a condition: a supply voltage, a temperature, a load, a frequency, a mode, sometimes a note reference that points to a paragraph three pages later. That column is not commentary. It is the domain over which the value is defined. Some of the traps it hides:
- Mode. "Deep sleep" on an ESP32 family part is several different states depending on which domains stay powered, and each has its own current. The row you want is the one whose conditions match the sleep configuration your firmware actually calls.
- Temperature. A leakage or quiescent-current figure at 25 °C tells you little about 45 °C, because leakage roughly doubles every 8–10 °C. This is the single most common cause of battery estimates that are wrong by a factor rather than a percentage.
- Load and duty. A sensor's "measurement current" is drawn during a measurement lasting milliseconds; its average contribution depends entirely on how often you measure, and the datasheet will often give a helpful "average current at 1 measurement per second" row that is useless if you measure once a minute.
- Supply voltage. Dropout, drive strength, and speed all move with the rail. A part characterised at 3.3 V behaves differently at the 3.0 V end of a discharging cell.
- What is not powered. Sleep current figures for microcontrollers usually exclude everything else on the board, and often exclude on-chip domains you need — RTC memory, a low-power timer, a retained GPIO hold.
When the graph beats the table
A table gives you a function evaluated at one or two points. The performance curves at the back give you the function. Four cases where the curve is the only honest source: quiescent current versus temperature (the table has 25 °C, your product does not live there); Rds(on) versus Vgs from Lesson 1.5 (the table has one gate voltage, you have another); capacitance versus DC bias from Lesson 1.4 (the table has zero volts, your circuit does not); and LDO dropout versus load current (the table has full load, your product idles at a few milliamps and the dropout there is far smaller — a rare case where the curve is good news). Where a vendor supplies a simulation tool instead of a curve, use it; where a vendor supplies neither, treat the parameter as uncharacterised and add margin accordingly.
SEE ITDatasheet claim matcher — pick the condition, watch the battery life move
Each control below is one row from a real datasheet, with the conditions that distinguish the options. Choose the one that applies to Beacon as the canon defines it. Every choice feeds a single additive current budget, and that budget becomes a battery life that either meets R1 or does not. The point is not the arithmetic — it is that a wrong reading of a conditions column is indistinguishable from a correct one until you divide.
Things to try. With everything at its canon setting the total is about 38 µA and the projection is comfortably past R1 — those are the canon's own numbers, and the largest single term is the 45 ms radio window, not sleep. Now change one control at a time and watch which ones matter. Switching the regulator to the XC6220 LDO — the part Beacon actually uses on the breadboard in Stages 1 to 3 — adds 25 µA, nearly doubling the budget from one component's quiescent current; that single row is why Lesson 4.2 exists. Switching the accelerometer to normal mode adds 9 µA for resolution Beacon does not use. Switching the radio to Wi-Fi multiplies the active term by nearly forty, which is the honest reason the canon advertises over BLE and only uses Wi-Fi for opt-in OTA. Then, with everything else correct, move the statistic control from typ to max over temperature and watch a three-year projection become something you would not put in a datasheet of your own — and notice that this is the one control where the "wrong" answer is the conservative one, which is exactly why design reviews ask for it.
EXAMPLEA guided tour of the three datasheets Beacon lives on
ESP32-C3-MINI-1 — the module, not the chip
The first thing to notice is that there are two documents. The module datasheet covers the ESP32-C3-MINI-1 as a soldered-down assembly: its pinout, its PCB antenna, its keep-out requirements, its certifications, and its recommended reflow profile. The chip datasheet and the technical reference manual cover the ESP32-C3 silicon inside it: peripherals, registers, electrical characteristics. Beacon buys the module, so the module datasheet is the contract — but almost every electrical number you want is in the chip datasheet, referenced from it. Both, plus the errata, plus the technical reference manual. Four documents for one part is normal for a microcontroller and unusual for anything else.
The rows that decide Beacon's design:
- Recommended operating supply: about 3.0 to 3.6 V. That range is the reason the regulator exists. A LiPo runs from 4.2 V down to 3.0 V, so the top of the cell's range is outside the part's recommended range and the bottom is exactly at its edge. Every regulator decision in Stage 4 is downstream of these two numbers.
- Deep-sleep current, on the order of 5 µA. Read the conditions: which power domains are retained, what wake sources are armed, and at what temperature. The canon's 8 µA total board budget allocates 5 µA to the module and the rest to everything else, which only holds while the firmware's sleep configuration matches the configuration the number was measured in.
- Transmit current. Given per mode and per output power, and hugely different between Wi-Fi and BLE. The canon's 40 mA average over a 45 ms window is a measured composite of a sensor read plus a three-packet BLE burst at 0 dBm, not a datasheet row — which is exactly the kind of number you have to measure yourself (Lesson 1.7).
- ADC characteristics. The ESP32-C3's SAR ADC is 12-bit with selectable input attenuation, and it is well known — documented by Espressif itself — to be non-linear and to need calibration, with per-chip calibration data programmed into eFuse. The datasheet gives recommended input voltage ranges per attenuation setting rather than a clean guaranteed accuracy, and there are documented constraints around using one of the ADC units while the radio is active. When a vendor phrases a specification as "recommended range" rather than min/max with an error bound, that is information: the parameter is not being guaranteed, and your design should not depend on its absolute accuracy. Beacon's battery reading is calibrated in production test (Lesson 6.3) for exactly this reason. Note also which pin the canon picks: battery sense is GPIO4 (ADC1_CH4), because ADC1 spans GPIO0–GPIO4 on this part and GPIO4 is in the RTC domain and is not a strapping pin. GPIO2 would have been electrically identical and functionally fatal, since it is a strapping pin that must read high at reset and the divider holds that node near ground with the PMOS off — and even with the PMOS on, 3.7 V halved is 1.85 V, below the 2.48 V that 0.75 × VDD requires. That constraint is not in the electrical characteristics table at all; it is in the boot-configuration chapter, which is the general lesson about where a datasheet hides the thing that will stop your board from starting.
- GPIO9 is a strapping pin. Not an electrical characteristic — a boot-mode input sampled at reset. The canon puts Beacon's only user button there, and the resulting trap is taught in 3.3 and recorded as a Rev A defect: holding the button at power-up confuses boot mode.
SHT40 — where "typical" meets a requirement
Beacon's R3 requires temperature to ±0.3 °C and humidity to ±3 %RH. The SHT40's high-accuracy grade quotes typical accuracies of roughly ±0.2 °C and ±1.8 %RH — comfortably inside the requirement, until you read how those are specified. The accuracy figures are typical, and they apply over a stated central portion of the sensor's range; the tolerance band widens toward the extremes of temperature and humidity. Beacon's environment (R9: indoor, 0–45 °C) sits in the good part of the range, which is why the part is a fit — but the correct statement in a requirements review is "typical accuracy meets R3 across our stated environment", not "the sensor is ±0.2 °C".
Three more rows matter. The measurement duration for high repeatability is a few milliseconds with a stated maximum — your driver's timeout must use the maximum, not the typical, or it will occasionally read a not-ready sensor. The measurement current of a few hundred microamps applies only during that window, so the average contribution at one reading per minute is a rounding error — the widget above makes that visible. And the sensor has an integrated heater with several power settings, intended for clearing condensation, which is also a warning: any self-heating, from the heater or from measuring too often or from a hot neighbour on the PCB, appears directly as a temperature error and an RH error. Sensirion's own application guidance is explicit about limiting measurement duty cycle for this reason, and it is why Stage 5's placement work keeps the SHT40 away from the regulator and, ideally, on a thermally isolated part of the board.
LIS2DH12 — a datasheet where the mode table is the whole story
The accelerometer's current consumption is not a number, it is a two-dimensional table indexed by output data rate and by operating mode, and it spans more than an order of magnitude. Low-power mode at a low ODR costs single-digit microamps; normal mode at a high ODR costs several times that. The trade is real: low-power mode gives 8-bit output data, while normal and high-resolution modes give 10 and 12 bits. Beacon needs to answer one question — "did this asset move since the last report?" — which is a threshold decision, not a measurement. Eight bits is plenty. Choosing low-power mode is therefore not a compromise, it is a correct match of the part's configuration to the requirement, and it saves most of the accelerometer's budget line.
The other row that matters is the interrupt configuration: R4 requires a report within 2 seconds of motion, and R3 requires motion wake within 100 ms, so the part's activity-detection threshold and duration registers, and the latency of its INT1 output, are requirements-bearing specifications. Those live in the register description rather than the electrical characteristics, which is a reminder that for a sensor the "datasheet" often includes an application note and a register map, and the numbers you need are spread across all three.
COMPARISONWhy the ESP32-C3-MINI-1 and not an ESP32-S3
The canon fixes the module. Here is the reasoning, made from datasheet rows and requirement numbers rather than preference — and with the honest counter-case at the end, because this is a decision reasonable engineers argue about.
| Dimension | ESP32-C3-MINI-1 | ESP32-S3 module | What Beacon needs |
|---|---|---|---|
| Core | Single RISC-V, 160 MHz | Dual Xtensa LX7, 240 MHz, vector extensions | Wake, read two I2C sensors, advertise, sleep. One core, and the single-core model makes the state machine of 3.7 simpler to reason about. |
| SRAM | 400 KB | 512 KB, plus PSRAM options | A 14-byte payload and two drivers. Not a constraint. |
| Radio | Wi-Fi 4 + BLE 5.0 | Wi-Fi 4 + BLE 5.0 | BLE advertising for reports (R2), Wi-Fi only for opt-in OTA (R7). Equivalent. |
| USB | Native USB Serial/JTAG | USB-OTG plus Serial/JTAG | Programming and debugging over one cable. The C3's Serial/JTAG is sufficient and removes a USB-UART bridge from the BOM. |
| GPIO count | Fewer, and several are strapping pins | Substantially more | Two I2C lines, one interrupt, one button, one LED, one divider enable, one LED-power enable. The C3 is tight but sufficient — and the tightness is what puts the button on GPIO9. |
| Module size and cost | Smaller and cheaper | Larger and more expensive | R8 caps the board at 35 × 25 mm; R5 caps the BOM at about USD 9.5 at 25 units, and the module is the largest single line. |
| Certification | Pre-certified module with a PCB antenna | Also pre-certified | Equivalent, and both remove the intentional-radiator certification burden discussed in 6.4. This is the single biggest reason to use a module rather than a bare chip. |
The S3 wins if the product grows. A second core lets a blocking I2C transaction coexist with a radio stack without restructuring the firmware; USB-OTG opens up acting as a host or a composite device; the extra GPIOs would have kept the button off a strapping pin and avoided one of Rev A's three known defects outright. If Beacon's roadmap included a display, an SD card, or on-device signal processing, the S3 would be the correct choice and the extra cost would be trivially justified. The C3 is chosen because the requirements as written (R1, R5, R8) are tight and the feature set is fixed — and the moment someone adds a requirement, this decision should be revisited rather than defended. That is what a requirements table is for, which is Lesson 1.9's subject.
BEACONEvery canon number, traced to a line and a condition
The discipline this lesson teaches becomes a document: for each number the design depends on, record where it came from, what conditions attach to it, and whether it is a guarantee or an estimate. Beacon's version of that table is below, and it is the artefact that makes Stage 4's power work auditable instead of hopeful.
| Beacon number | Source | Statistic | Conditions that must hold |
|---|---|---|---|
| Module deep sleep ≈ 5 µA | ESP32-C3 chip datasheet | typ | Specific sleep mode, retained domains, wake sources armed, 25 °C. Verify against the firmware's actual sleep call and measure at temperature. |
| Board deep sleep ≈ 8 µA | Beacon canon — measured | measured | Module + buck IQ + sensors idle + divider isolated + leakage. Measured on hardware (1.7), not summed from datasheets. |
| Active window ≈ 45 ms at 40 mA | Beacon canon — measured | measured | Sensor sample plus a three-packet BLE burst at 0 dBm. A composite of many datasheet rows, so it must be measured, not derived. |
| Average ≈ 38 µA | Derived: 8 µA + 40 mA × 45 ms / 60 s | derived | One report per minute (R1's condition). Any change to the report interval invalidates it. |
| ≈ 3 years theoretical life | Derived: 1000 mAh / 38 µA | derived, nameplate capacity | Nameplate capacity, no self-discharge, no temperature effect, no retries. This is the optimistic bound, and the canon says so. |
| ≥ 6 months guaranteed (R1) | Requirement, canon | requirement | Holds even with a 6× real-world derating against the theoretical figure. The margin is the design's actual safety factor. |
| Buck IQ 60 nA | TPS62840 datasheet | typ | A specific mode and no load. Check the curve against temperature; 60 nA is small enough that even a 5× error is invisible against 8 µA. |
| LDO supply current 25 µA typ | XC6220 datasheet | typ | Stage 1–3 breadboard only. Three times Beacon's entire sleep budget on its own, which is the whole argument of Lesson 4.2. |
| Temp ±0.3 °C, RH ±3 % (R3) | Requirement vs SHT40 typ accuracy | requirement vs typ | SHT40 typicals meet it inside Beacon's 0–45 °C environment; the band widens outside. Self-heating must be avoided by duty cycle and placement. |
Notice the third column. Three of Beacon's most load-bearing numbers are measured, not taken from a datasheet — the board sleep current, the active window, and the composite active current. That is not a failure of the datasheets; it is the correct division of labour. A datasheet tells you about one component under stated conditions. Your board is many components under your conditions, and no document describes it. The engineering skill is knowing which numbers you can read and which you must measure, and the reliable test is: if the number depends on more than one part, or on your firmware, it must be measured. Lesson 1.7 is about doing that measurement without the instrument changing the answer.
WEAKNESSWhere this bites
Every term in such a sum is the centre of a distribution measured at 25 °C with everything else off. The sum of typicals is not a typical of the sum in any useful sense, because the terms are correlated — temperature moves all of them the same way, and so does supply voltage. A budget built this way is systematically optimistic, and it is optimistic by a factor rather than a percentage. Do it anyway, because it tells you where the effort should go and it costs an hour. Then treat the result as a hypothesis, measure the real board (1.7), and put the measured number in the requirements table. Beacon's canon does exactly this: 8 µA and 45 ms at 40 mA are measured composites, and the derived 38 µA and 3 years are labelled as derived.
Look for the footnote markers. Parameters marked "guaranteed by design", "by characterisation", or "not production tested" are statistical claims about a process, not screens on the unit you received. Parameters given only as a typical with no min or max are not guaranteed at all. Neither of these is a scandal — it is often uneconomical to test a leakage current on every unit — but it changes what you may build on top of them. The practical rule: if your design's margin on a parameter is less than 2×, and that parameter is a typical-only number, you either need to test it in production yourself, choose a part that guarantees it, or restructure so the margin does not matter. Beacon's ADC accuracy takes the third route by calibrating in production; its sleep current takes the first by measuring on a jig (6.3).
Three distinct versions of this. Wrong revision: parameters move between revisions, and a distributor's cached PDF can be years old — always fetch from the manufacturer and check the revision date against your copy. Wrong grade: ordering information often lists several part numbers whose only difference is temperature range, accuracy grade, or tolerance, and the electrical characteristics table sometimes covers all of them with grade-dependent footnotes. The SHT40 has multiple accuracy grades; buying the wrong one silently changes whether R3 is met. Wrong document: for a module you need the module datasheet, the chip datasheet, the technical reference manual, and the errata, and the number you want is rarely in the first one you open. Keep the revision and date of every datasheet in your design notes; when a board behaves unexpectedly two years later, that record is how you find out whether the part changed under you.
KEY INSIGHTHow this connects
- What an Embedded System Is (1.1) — the constraints that define embedded work are almost all datasheet numbers with conditions attached, and this lesson is how you read them without inheriting someone else's optimism.
- Voltage, Current, Resistance, and Power (1.2) — the power rating and its derating curve were the first example of a number that is only valid at a stated ambient. Every parameter in this lesson behaves the same way.
- Passive Components for Real (1.4) — capacitance at 0 V bias and inductor Isat at an unstated percentage drop are the conditions column doing its worst, on components most people never read a datasheet for.
- Diodes, Transistors, and MOSFETs as Switches (1.5) — Rds(on) at Vgs = 10 V and a threshold spread of 2:1 are the same lesson applied to a switch, with a thermal failure at the end of it.
- Measurement (1.7) — the next lesson covers the numbers you cannot read and must measure, and the ways your instrument changes them while you look.
- Beacon's Requirements (1.9) — a requirement with units and conditions is a datasheet row you write for yourself, and it is judged by exactly the standards in this lesson.
- Sleep Modes and the Power Budget (4.3) — where the budget assembled in this lesson's widget becomes a measured, defended number on real hardware.
Of every skill in this course, this is the one with the highest return per hour invested. Circuit theory transfers to circuits; firmware transfers to firmware; but reading a specification with its conditions attached transfers to every part, every standard, every API, and every contract you will ever read. The habit is small: never write a number in a design document without also writing where it came from and what has to be true for it to hold.
GLOSSARYKey terms introduced
- Absolute maximum ratings
- Stress limits beyond which permanent damage may occur. Nothing is guaranteed to function at them, they generally may not be applied simultaneously, and exposure can degrade reliability latently. They bound your protection design, not your operating design.
- Recommended operating conditions
- The envelope within which every other guarantee in the datasheet is valid. Your design must live entirely inside it, at every temperature and supply voltage it will see.
- Typ
- The centre of a characterisation distribution, usually at 25 °C and nominal supply, and frequently not production tested. Useful for estimating a population; not a promise about your unit or your temperature.
- Min / max
- The guaranteed limits, usually across the full recommended operating range. The only numbers you can build a design margin on — subject to any footnote saying they are guaranteed by design rather than tested.
- Conditions column
- The supply voltage, temperature, load, mode, and duty cycle under which a specification holds. The other half of every number; a value quoted without it is not a specification.
- Guaranteed by design / by characterisation
- A footnote meaning the limit is derived from process knowledge or a sample, not screened on every unit. A weaker promise than a tested limit, and a reason to add margin.
- Errata
- A separate document listing known silicon defects and their workarounds, indexed by chip revision. Rarely inside the datasheet, and the first thing to read when a peripheral misbehaves.
- Derived number
- A figure computed from other figures rather than read or measured. Must carry its inputs and their conditions, because it inherits every assumption in them — Beacon's 38 µA average and 3-year life are both derived.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1A part's absolute maximum supply is 3.9 V and its recommended operating range is 3.0–3.6 V. Your regulator's output tolerance is 3.3 V ±5 %, and its input is a LiPo that reaches 4.2 V. Which numbers actually constrain your design, and what would you check?›
The recommended range constrains the design; the absolute maximum constrains the protection. Regulator output at ±5 % spans 3.135 to 3.465 V, comfortably inside 3.0–3.6 V, so the steady state is fine. Then check the cases the steady-state number hides. Transients: a load step from the BLE burst causes overshoot and undershoot on the rail; if the overshoot pushes past 3.6 V the part is momentarily outside its guarantees, and if it pushes past 3.9 V you are risking damage. Start-up: what does the rail do before regulation, and what does it do if the battery is connected while the regulator is enabled? Fault: if the regulator fails short, the 4.2 V cell appears directly on a 3.9 V absolute maximum — which is a good argument for a buck whose failure modes you have thought about, and for the TVS and protection work of Lesson 2.6. Bottom of range: at the empty end, an LDO with 200 mV of dropout from a 3.0 V cell cannot produce 3.0 V at the part's pin, so the low end of the recommended range and the cell's cutoff voltage interact — which is the argument for the buck.
2You are given only a typical value for a microcontroller's deep-sleep current, with no maximum. Your budget has 2.5× margin on that term. Is that enough? What are your options?›
Probably not, and the reason is temperature rather than unit-to-unit spread. Sleep current in a modern microcontroller is dominated by leakage, which roughly doubles every 8–10 °C. Beacon's R9 allows 45 °C, which is 20 °C above the 25 °C the typical was measured at — a factor of roughly 3 to 4 on the leakage-dominated portion, before any unit-to-unit variation. A 2.5× margin is consumed by temperature alone and leaves nothing for the distribution. Four options, in order of preference. (a) Measure it on real hardware at the top of your temperature range — the only option that produces a defensible number, and it is what Beacon's canon does for the board's 8 µA. (b) Find the curve: many vendors publish quiescent current versus temperature even when the table gives only 25 °C. (c) Restructure so the margin does not matter — if sleep is 8 µA of a 38 µA average, a 4× error on sleep takes the average to 62 µA and R1 still passes, which is exactly Beacon's situation and the reason the design is robust. (d) Choose a part that guarantees a maximum. Note that (c) is the strongest answer here: the best defence against an uncertain parameter is a design in which that parameter is not the dominant term.
3Beacon's R3 requires ±0.3 °C. The SHT40's datasheet gives a typical accuracy of about ±0.2 °C. Is R3 met? Write the sentence you would put in a design review.›
Not as stated — the comparison is between a requirement, which must hold, and a typical, which describes a population at stated conditions. Three things must be established before the claim is defensible. Grade: the SHT40 is sold in more than one accuracy grade, so the requirement is on a specific ordering part number, not on "SHT40". Range: the accuracy figure applies over a stated central portion of the temperature and humidity range and widens outside it; Beacon's R9 environment of 0–45 °C sits inside the good region, which must be checked and stated rather than assumed. System error: the sensor's own accuracy is not the system's accuracy. Self-heating from the measurement itself, from a nearby regulator, and from the enclosure adds error in one direction, and it is a placement and duty-cycle problem rather than a sensor problem. A defensible review sentence: "R3's ±0.3 °C is met by the SHT40 high-accuracy grade, whose typical accuracy of ±0.2 °C applies across our 0–45 °C environment, provided self-heating is held below 0.1 °C by limiting measurement to one high-repeatability reading per minute and keeping the sensor thermally isolated from the regulator — both of which are verified in Stage 6 bring-up." Notice that the sentence names the grade, the condition, the system-level error term, and the verification method.
4A colleague's battery-life estimate for a similar product comes out four times better than yours for the same hardware. Neither of you has made an arithmetic error. List the four most likely places the difference is hiding, in order.›
(1) Sleep mode. The largest single lever, and the easiest to read wrong: deep sleep and light sleep on this class of part differ by more than an order of magnitude, and "sleep" in a conversation does not distinguish them. Ask which API call the firmware makes. (2) Duty cycle of the active window. The radio term dominates the budget, and it scales linearly with both burst length and report interval. A budget assuming one report every five minutes beats one assuming every minute by 5× on that term alone. (3) Statistic and temperature. Typical at 25 °C versus maximum over the operating range is comfortably a factor of two to three on the leakage-dominated terms. (4) What is counted as the battery. Nameplate capacity versus usable capacity above the cutoff voltage, with or without self-discharge and ageing, is worth 20–30 % — smaller than the others, and the one people argue about most. Work through them in that order, because the first two are usually the whole difference and the last two are where the discussion goes if you start at the wrong end. This is also the argument for writing conditions next to every number in a budget: the disagreement becomes visible in seconds rather than in a meeting.