Differential Power Analysis
Differential power analysis (DPA) extracts a chip's secret cryptographic key by measuring how much current it draws while it computes, without ever touching the logic itself or breaking any encryption mathematically — the attack works because the power a circuit consumes depends on the data it's processing, and that dependency leaks information about secret bits one statistical correlation at a time. Published by Paul Kocher, Joshua Jaffe, and Benjamin Jun in 1999, DPA turned a side effect every chip designer had always known about — power draw varies with the data being processed — into a practical way to pull an AES or RSA key straight out of a smart card or secure microcontroller using nothing more than an oscilloscope and statistics.
The attack never needs to understand what any individual trace shows — it needs hundreds or thousands of them. A single power trace is far too noisy to reveal anything about the key by itself, but DPA's trick is to guess one bit of the key, split every recorded trace into two groups based on what that guessed bit predicts the circuit's internal activity should have been, and then average each group — if the guess was correct, the averaging cancels out everything unrelated to that bit and leaves behind a small but real spike exactly where that bit's effect occurs in time; if the guess was wrong, both groups average out to the same noise with nothing distinguishing them.
The target never has to be physically opened, which is what made DPA so alarming when it was published. Earlier invasive attacks required decapsulating a chip and probing it directly, a destructive process requiring real lab equipment and sample preparation; DPA only needs the device to be powered and operating normally while an attacker measures the current it draws from the outside, which meant a huge population of already-deployed smart cards, pay-TV decoders, and secure microcontrollers were suddenly vulnerable to key extraction using equipment far cheaper than anyone had assumed was necessary.
Every real countermeasure works by breaking the correlation between data and power, not by hiding the power signal itself. Masking randomizes intermediate values with a random blinding factor so the power trace no longer correlates with the real secret data at any single point, hiding techniques add random timing jitter or balance circuit activity so every operation draws roughly the same current regardless of the data, and dedicated secure-element and smart-card chip vendors such as Infineon, NXP, and Thales build DPA resistance directly into their cryptographic cores specifically because the attack is cheap enough that any product handling payment or identity credentials has to assume someone will eventually try it.
Read differential power analysis through a *statistics-over-secrecy* lens rather than a *code-breaking* lens — nothing about the encryption math was ever broken, the chip simply couldn't stop its own power consumption from correlating with the secret it was trying to protect, and once that correlation exists in enough traces, averaging alone is sufficient to pull it back out.