n && n == (n & (~n + 1))
One's complement machines will only work with the above expression. If you're only working with two's complement machines where negation may take the same or fewer clock cycles, assuming that even matters, it can be shortened to the following:
n && n == (n & -n)
I'd create a C program iterating -64 to +64 inclusive that prints the binary values as well as decimal to see why this works.