A datasheet is a contract, not a brochure
You spent four guides turning the ideal op-amp back into a real one — discovering its DC errors, its finite bandwidth and slew rate, and the way its feedback loop can break into oscillation. Every one of those imperfections is written down, as an exact number, in the part's datasheet. Learning to read it is the capstone of the rung: it is where everything you understand turns into a part you can actually buy and trust. Think of it as the manufacturer's signed contract — they promise the chip behaves within these bounds, and nothing more.
Almost every op-amp datasheet has the same three load-bearing sections. First, absolute maximum ratings — the do-not-exceed page (max supply voltage, max input voltage, junction temperature); cross any of these and the part is allowed to die. Second, the electrical characteristics table — the guaranteed numbers, each measured under stated conditions. Third, the typical performance curves — graphs of how those numbers slide with supply, temperature, and frequency. The table is the spec that matters most; everything else is context for reading it correctly.
The DC error block
The top rows of the table are the DC errors from guide 2 of this rung. First is the input offset voltage Vos — the tiny mismatch voltage you would have to apply between the inputs to make the output sit exactly at zero. Your circuit multiplies it by the full closed-loop gain, just like a real signal. With a 3 mV max Vos in a non-inverting amp of gain 100, the output carries up to 3 mV times 100 = 300 mV of pure DC error, present before any signal arrives. Right beneath Vos sits its drift, in uV/C: multiply it by your temperature span to see how much that offset wanders as the board warms up.
The next two rows are the input currents. input bias current Ib is the small steady current each input actually draws (the golden rule's 'no input current' is an idealization). It becomes an error voltage when it flows through your resistors: 100 nA through a 100 k feedback resistor makes 100 times 10^-9 times 100 times 10^3 = 10 mV of offset. The classic fix is to match the resistance seen by each input so Ib cancels — leaving only the input offset current Ios, the difference between the two bias currents, which is usually ten times smaller. That is why those two rows always appear as a pair.
Be honest about scale, though: these DC errors only matter relative to your signal. For a 2 V audio waveform, a 1 mV offset is a rounding error you will never hear. For a thermocouple delivering 40 uV per degree, that same 1 mV is twenty-five degrees of pure lie — there you must reach for a precision part with sub-100 uV Vos and single-digit uV/C drift. The datasheet does not tell you whether the numbers are good; it tells you what the numbers are, and your application decides.
The AC block: gain, bandwidth, slew rate
Now the AC rows from guides 1 and 3. The open-loop gain Aol is quoted in decibels — 120 dB means a raw gain of 1,000,000. That huge number is your loop gain reserve: the more of it you spend on feedback, the more precise and linear your closed-loop amplifier becomes. But it is only that big at DC and falls steadily with frequency, which is exactly what the gain-bandwidth product (GBW) captures. The closed-loop bandwidth is simply GBW divided by your gain: a 10 MHz part set for a gain of 100 runs out of bandwidth at 10 times 10^6 / 100 = 100 kHz. Want more gain and more bandwidth? You must buy a part with a bigger GBW.
The slew rate is a separate limit, and beginners constantly conflate it with bandwidth. Bandwidth is about small signals; slew rate is the fastest the output can ramp, in V/us, no matter how hard the input shoves. A big swing at high frequency hits it first: the full-power bandwidth is fp = SR / (2 times pi times Vpeak). A fine 20 V/us part trying to make a clean 10 V-peak sine can only do so up to 20 times 10^6 / (2 times pi times 10) ≈ 318 kHz — far below its small-signal bandwidth. Push past that and a sine wave degrades into a triangle. Finally, the output swing rows tell you how close to each supply rail the output can actually reach; a rail-to-rail output gets within millivolts, an older part may stop a volt or more short.
ELECTRICAL CHARACTERISTICS (Vs = +-15 V, Ta = 25 C, unless noted)
--------------------------------------------------------------------
Parameter Symbol Min Typ Max Unit
--------------------------------------------------------------------
Input offset voltage Vos -- 0.5 3.0 mV
...vs temperature -- -- 2 10 uV/C
Input bias current Ib -- 20 100 nA
Input offset current Ios -- 2 25 nA
Open-loop gain Aol 100 120 -- dB
Gain-bandwidth product GBW -- 10 -- MHz
Slew rate SR -- 20 -- V/us
CMRR CMRR 80 100 -- dB
Input voltage noise en -- 8 -- nV/sqrt(Hz)
Output swing (RL = 10k) Vo +-13 +-13.5 -- V
--------------------------------------------------------------------
Read Min/Max as the guarantee. Typ is the promise nobody signs.Rejection and noise
Two rows decide whether you can pull a small signal out of a messy world. The common-mode rejection ratio (CMRR) measures how well the op-amp ignores a voltage applied equally to both inputs — the hum and interference that ride on both wires together. A 100 dB CMRR means a 1 V common-mode disturbance leaks through as if it were only 1 / 100,000 = 10 uV of real input. When the signal you want is a tiny difference sitting on a large shared voltage — a bridge sensor, a current-shunt — you lean on this hard, which is exactly why the instrumentation amplifier is built around maximizing it.
The noise row is quoted as a density, like 8 nV/sqrt(Hz) — voltage per square-root of bandwidth, because broadband thermal noise piles up with how wide a band you let through. To get the actual RMS noise, multiply by the square root of your bandwidth: 8 nV/sqrt(Hz) times sqrt(100,000 Hz) ≈ 2.5 uV RMS over a 100 kHz band. Down at low frequencies the density rises — that is flicker (1/f) noise, and the frequency where it overtakes the flat part is the '1/f corner', a number that matters intensely for slow, precise DC measurements. All of it sets the noise floor: the smallest signal that is not simply drowned.
Stability fine print, and choosing the part
Guide 4 warned that feedback can bite back, and the datasheet hides the clues. Look for the words 'unity-gain stable' — they promise the part stays calm even at a gain of 1, with healthy phase margin. Then hunt for the maximum capacitive load it can drive: a long cable or an ADC sampling cap adds capacitance that erodes phase margin and can set the loop ringing or oscillating. The cure is usually a small series resistor to isolate the load, or a part with built-in compensation. None of this screams from the front page — it lurks in a stability graph and a one-line note.
Always come back to the absolute maximum page before you commit. A part rated for a 36 V total supply should never be operated there — you leave headroom and derate, running well inside every limit so heat, aging, and tolerance never push you over the edge. Good design lives in the comfortable middle of the ratings, not at their cliff. With that, you are ready to turn the whole datasheet into a decision.
- Start from YOUR spec, not the part. Write down the signal: the smallest voltage you must resolve, the closed-loop gain you need, the highest frequency, the supply you will run on, and the load you must drive. This is component selection working backwards from the job.
- Build a DC error budget. Add Vos, plus Ib times your resistance, plus drift times your temperature span; multiply by the gain and compare to your smallest signal. If the error swamps the signal, you need a precision (low-Vos, low-drift) part.
- Check the AC budget. Require GBW greater than gain times your highest frequency, AND a slew rate fast enough for full-amplitude output: fp = SR / (2 times pi times Vpeak) must clear your top frequency at full swing.
- Confirm output and supply. Will it swing far enough into the rails and source enough load current? On a single low supply, you probably need rail-to-rail inputs and/or output.
- Weigh noise, rejection, stability — then derate. For tiny signals check noise density, the 1/f corner, and CMRR. Confirm it tolerates your load capacitance. Step back inside the absolute-max ratings, and the part is yours. Every step is a tradeoff; the datasheet just makes them honest.