I made it using a print that I developed for other purposes (with display, few LEDs and keys). The processor on the print is AT89S8252, so I use the built-in SPI. The print has a big RAM (relatively - 128k), where I first download the program to be programmed.

Of course, I follow all the recommended timings and sequences. After writing to SPDR, I make polling of bit 7 of SPSR (SPIF bit), and then make a small delay (ACC=60, DJNZ ACC).

In case of S8252/S53 I verified each byte immediately, with number of tries limited to 12. So, as soon as the byte is verified, it goes on to program the next byte. The program adapted itself automatically to the one or the other processor, depending on the code is less or greater than 8kb (there is a small difference in instruction bytes).

With AT89S8253, I chose, of course, the 64 byte blocks programming. After programming a block, and 4 ms pause, I start verifying the block. If any byte mismatches, the flag is set (you have to continue to read the entire block). Number of tries: 3, although 2 is enough.

When it finally succeeded, it showed a big advance in programming speed (haven't yet measured, but I suppose at least 5 times faster).

Regards, Slobodan