On 7/26/05, Tony Duell <ard at p850ug1.demon.co.uk> wrote:
Remember the interface is
bit serial, preesumably your memory is at least 8 bits wide, you may well
have to start writing in the middle of a memory word.
Are you sure about that 50 MHz? I would have guessed 10 MHz would be
fast enough.
Um, wouldn't you use each bit in a word as a different head rather
than than address bits withing a word? It's not like you can't
perform an SRAM read+write cycle in 20 ns, although direct to [S]DRAM
might be a bit more challenging. So in terms of 50 MHz sampling,
you're talking 1 Mbit per head of fast memory for cylinder cache (you
might as well choose a multiple of 8 heads), 1Mbit per head per
cylinder of slow non-volatile memory for storage, and a PIC to
transfer between the two (handle the "stepping") and a FPGA to hold
the logic (a resettable 50 MHz, 20 bit counter tristated to the cache
address bus and the read/write logic) So a 20 MB drive (8*640*17) is
1 MByte of fast memory, and 640 MB CF card. If 10 MHz is fast enough
you can reduce those memory requirements by a factor of 5.
As I said, it's not that difficult, just a bit
harder than you may have
first thoguth.
Sounds like fun. :) Wish I had the time...
Eric