*** M E M O R U M ***

If anyone remembers Memofest 2018, there was a strange 8 pin chip that found it's way into the photo of the NFX network card.

I think I mentioned at the time that it was an 8 meg serial flash, and capable of taking an entire CPM partition.

I've now managed to get something build to test the idea of running CPM from an SPI flash.

Since 2018, the flash chip I was playing with has gone out of production, the modern replacement is actually twice the capacity, so that's now 16 meg of storage in a 6mm square package, enough for 2 partitions. There's 2 sockets on the board, so in theory there could be 4 partitions, but in practice there isn't enough software to warrant using a 2nd device. There are larger devices available that would fit 4 partitions or more on a tiny chip, but the data transfer protocol is different and again there's no reason to make the change unless the current chip goes out of produsction.

The chip is not only serial, it's 3 volt CMOS. So as well as converting parallel to serial to parallel, the signals have to be level shifted between 3.3v and 5v.
The test board is double ended, allowing the early work to be done on the MTX plus. Once the I/O routines were running well, the ROM could be written, and testing moved to the MTX512.

Working from the bottom up there is a CPM boot rom and 20 pin GAL to decode the I/O port. The board uses a single port (128, #80) for all transfers. Above the rom are 3 LED's for one for each chip select and one for the SPI clock of an activity indicator. There is a 3 volt regulator to power the flash and 3v devices and some jumpers to account for the differences between the MTX and the MTX plus.

The 20 pin chip in the middle is a 74HC273 octal flip flop which is the output port and runs at 5v. My test MTX has a CMOS CPU for a stock MTX an HCT273 would be needed. The 273 has a clear input attached to reset so is in a known state at start up.

The top row of chips is a 74HC04 hex inverter, it's running from the 3v supply so has to be HC or some other 3v family device. It's providing the output for all 3 LEDs as well as the flash chip selects and the 3v data output port enable. The output port resets to zero, the inverters ensure the flash chips are inactive (high), as is the write buffer.

In the middle is a 74LVC244. I picked the LVC version as it's a 3 volt device but has 5v tolerant inputs. It can take the 5 signals from the output port, and convert them to 3v for the flash. In addition it has 2 separate 4 ports each with their own enable. The 3 state output is something the 273 lacks. One 4 bit port is permanently active, that supplies the 2 chip selects and clock to the inverter and also the enable for the other 4 bit port. The 2nd port deals with the data transfer to the flash, and needs to be 3 stated when the flash is outputting data

The final 20 pin chip is a 74HCT373 octal latch, it's run in transparent mode, with the IN (#80) decode attached to the output enable. 5 of the 8 bits are used, the other 3 read low. The flash chip is configured to supply data in 4 bit mode, however it can also use 1 bit mode. For that reason the while lower 4 bits mirror the 4 bit output, the output that would be used in single bit mode is duplicated at bit 7 for ease of access.

The 2 empty sockets at the top are where the decode GAL goes on the MTXplus and the unused device 2 socket. There are no shift registers or anything like that, the entire serial protocol is handled in software.
In place on the MTX it illustrates how impractical vertical add-ons are on the MTX. The case end panels have been removed and the keyboard section of the case offset to avoid touching anything on the board. I didn't have any spare right angle edge connectors so it had to be that way.
The CPM rom is just the one I put together for the NFX board, with the "disc" driver software re-written. For "Day to Day" use, a custom loaded partition would be more use.

I've just dropped 2 partitions from my Memu install onto the flash for testing, One with the 59k system, the other has the 54k version.

Speed wise it's not as fast as CFX or CFX2 because it's transferring 4 bits at a time instead of 1 or 2 bytes, however it's still capable of over 30k a second so for the size of the usual MTX games load times aren't an issue. There is also some tweaking room in the driver and deblocking routines to speed things up a little.

Both SPI flash chips I tried have 4k sectors, which would seriously compromise write access as all 4k would need to be written each time CPM write a 128 byte "sector" in addition 4k of ram would have to have been found to buffer the sector. Unlike the CFX or Andy's ReMemorizoer (which is also SPI) where the sector size is a more reasonable 512 bytes. Which is why it's read only, a writeable CPM system would be seriously slow.

An alternative version of the boot rom with the SDX basic extensions instead of CPM would be pretty straightforward to produce, as only minor tweaks to the low level driver source would be needed.

The current source for the SPI driver:

If the chip is supplying 4 bits of data, would it be better to have two chips in parallel to supply 8 bits? Would the speed improvement in not having to repack the data be worth the cost of the extra chip?

Interesting idea.

The chips are pretty cheap, of the order of £1.20 each in 10's. Putting 2 on won't break the bank! (They are much, MUCH cheaper than parallel flash which costs about that for 1/2 a meg, and imagine the board size and decoding required for 32 of those!)

Adding the extra output port hardware would cost more than the flash would. Decoding 2 IO ports, one for 8 bit data one for control isn't a problem the decode gal has 5 spare pins at the moment.

But for a read only system that might not be needed, the only thing being sent to the flash is the read data command, and that's sent 1 bit at a time using the serial protocol. Keeping the output from the 2 serial in pins from interfering with each other is the only connection issue, and that can probably be sorted by re-jigging the output port buffer.

Because everything is clocked at around 1% of the maximum speed of the chip, any timing difference between devices shouldn't be an issue.

Splitting the Partition image(s) into 4 bit wide data shouldn't be too hard, a bit of Basic on the RiscPC would deal with that.

Having said that, the current system was designed with the the Z80 RLD instruction in mind, the repacking isn't a chore. What 8 bit transfer would save is the entire 2nd clocking sequence, and that's actually slower than the first half because RLD is much slower than LD (HL),A.

I'm tempted to give it a try.