How does ROM work? Read-only memory explained

ROM, read-only memory, takes an address and gives back the word stored there. Inside, a decoder lights one word line per address, and each output bit is an OR of the word lines whose word has a 1 in that bit. The pattern of connections is the data. It's fixed when the chip is made, so it's still there after the power goes off.

Decoder first

A 2-bit address picks one of four word lines. Line 0 is NOT A1 AND NOT A0, line 1 is NOT A1 AND A0, line 2 is A1 AND NOT A0, and line 3 is A1 AND A0. Exactly one is lit at a time. An address of n bits reaches 2ⁿ words: 3 bits, 8 words; 16 bits, 65,536.

ORs make the bits

Store 5, 3, 6 and 0 at addresses 0 to 3:

5 is 101, so line 0 joins the 4s OR and the 1s OR. 3 is 011: the 2s and the 1s. 6 is 110: the 4s and the 2s. Address 3 holds 0, so its line joins nothing, and when it's lit every OR gives 0. Changing the data changes only the joins, never the decoder.

A grid of diodes

Early ROMs were literally this grid: word lines crossing bit lines, with a diode soldered at every crossing where the word has a 1. Diodes from several word lines into one bit line make a diode OR: whichever line is high pulls the bit high, and the diodes stop it pulling the other lines up. Mask ROMs made the same grid in silicon, with the pattern set by one of the chip's last layers.

Fetch: a counter on the address

Put a counter on the address lines and every tick reads the next word: 5, 3, 6, 0, and round again. That's the fetch step of every processor, with the program counter addressing program memory and the word that comes back being the next instruction.

From ROM to flash

EPROMs could be erased with ultraviolet light through a quartz window in the chip, EEPROMs electrically, a byte at a time. Flash is their descendant: read like a ROM, rewritten in blocks. An Arduino's sketch lives in flash, which Upload rewrites, and old game cartridges were plain mask ROMs, which is why the games never changed.