Floating point already does that with separate mantissa and exponent bits. Regardless, you still have a limited number of bits for precision.