Why an ARM processor with 32 bits address bus can address 4 billion different bytes?

Why an ARM processor with 32 bits address bus can address 4 billion different bytes? I know that $2^{32}$ is equal to about 4 billions, but shouldn't it be 4 billion bits and not bytes?

Hence if I want to find how many different words it can address it should be $\frac{2^{32}}{8}$ so I will have the number of bytes, and if I want to find the number of words I divide it again by 4 (because a word in ARM is 32 bits = 4 bytes).

• ARM (like most modern ISAs) uses byte addressing. (ARM does have a feature to allow bit-level addressing, mapping a section of the address space (bit band) for this. Such is primarily intended for memory-mapped I/O where atomic bit-level addressing can be useful since accesses to I/O devices can have side effects (so a byte read to get a single bit could generate unintended side effects for the seven other bits read and a read-modify-write sequence to change one bit could have unintended effects).) Dec 30 '16 at 14:52

This means that each possible address corresponds to a potential byte in memory. Thus a 32 bit address would have a span of $2^{32}$ or $4,294,967,296$ bytes.