A 16MHz AVR like the one in an Arduino can read an 8-bit port fairly frequently, in theory every 4 cycles or so (4MHz), though I think in a real program you’d be lucky to get every 15 cycles (1MHz). It occurred to me that you could maybe do a quasi-equivalent of an R-2R DAC to get some kind of sample at that frequency: by running different voltage dividers from a voltage source across the different pins, the digital level thresholding can maybe give you some approximate idea of how high a signal is: a 3-bit idea of it, to be exact.
This sounds kind of useless, but I was thinking you could use it in combination with the ADC, which on the ATMega328 runs at up to 76.9 ksps or up to 15 ksps at maximum resolution. Each conversion takes 13 ADC clock cycles, where the ADC clock runs at between 1/128 and 1/2 the system clock, thus 125kHz to 8MHz at a 16MHz system clock speed; this gives us 9.6 ksps to 615 ksps, although I’m guessing that running the ADC clock at over 1MHz won’t produce useful results.
I was thinking that you could run the normal ADC at its 15 ksps speed, thus providing an absolute 10-bit measurement of the input voltage, but then sample a high-pass-filtered version of the input voltage at about 1MHz and only three bits per sample. This would give you about 64 3-bit relative samples for each 10-bit absolute sample, gathering information about frequency components up to 500kHz. If there is significant overlap in the frequency responses of the two parts of the system, you can use the slow, absolute, high-precision ADC to calibrate the fast, relative, low-precision ADC.
You probably want to buffer the input voltage on its way into the high-pass filter and voltage dividers in some way — at least an emitter follower or source follower, maybe a whole op-amp buffer or something.
This seems like it might be capable of converting an ATMega328 into a 500kHz-bandwidth “oscilloscope”, which is like 2½% of a real oscilloscope.