Call for final Elf99 design input

From: Ethan Dicks <erd_at_infinet.com>
Date: Wed Nov 25 15:07:10 1998

>
> > o EPROM enable switch.
>
> Where is the original PROM ?

Since I never used it (an 82S123), I'm not certain. It seems to map into
$0000 - $001F on the original Elf when the PROM switch is set to enable.

> Use a part of the EPROM constantly maped at $0000 or add an
> 1824 32x8 RAM ?

I don't get this.
 
> > o Circuit to ghost EPROM at $0000 until first address access at $8000
>
> I still would go for a better decoding - just the high bit is to short.
> Maybe there are some spare gates to use ?

Wait a minute. We have been having an extensive offline conversation
and you said that 32K of RAM and 32K of ROM is enough. It can be more or less
of either, at a cost of gates and complexity.

Going over the Quest design, I managed to eliminate a 4049 by redesigning
the PROM area - the original used several inverters and NANDs to decode A6-A8.

> And I still would like to combine the shadow and boot option.

If I did not convey that properly, I'm sorry. Yes, the boot option
is always there because it's software. The hardware echoes the ROM at
$0000 and $8000 when the PROM enable switch is on, and the act of reading
a valid ROM address (high memory) resets the flip flop, restoring the
more normal RAM at $0000 and ROM at $8000. If you make the first two
instructions INP 4 and PHI 0 (your suggestion), you get the ROM bank select
feature that you originally proposed.

> > o Sockets for two 1822/2101 256x4 SRAMs, the same style used by
> > the original Elf.
>
> Esential to be as close as possible to the original design.

One vote for, several against or abstaining.
 
> > o A 1851 programmable input/output port -or- a pair of 4508 latches
> > wired as one input and one output port (as in the COSMAC VIP)
>
> I would go for two 1852 Ports. They use only one adress each.

I'm not inclined to do dual 1852's. We you and I have discussed, the 1851
uses three ports, the switch/display is another port, leaving three left
over. While ports are scarce, they are no so scarce as to overshadow the
benefits of a bit-programmable I/O port.

The space taken up by two 40 pin chips is a bit much when I can implement
two cheap ports (one in and one out) with octal latches.

> > o A 1854 UART
>
> Data transfer via Q is way more fun (Hi Alison :).
> And since we don't need high speed transfer (2400 is ridicoulous i
> we can do 110 :) a complete software solution is a great thing to do.

I have no problems in principle with a software UART, but I do have
something to say about the speed. 2400 baud is *not* a ridiculous
speed, especially if you want to talk to an external device that has
a fixed clock, say, a serial-to-LCD board. The other issue with a
software UART is timing. If you use a hardware UART, it needs a
crystal of a particular frequency, but the CPU does not. You can then
"clock chip" the 1802 up (to use a Mac term) and not recode your serial
routines.

> > o Either a dual 4042 address latch or a single 4508 latch. One or
> > the other is needed to implement more than 256 bytes of memory.
> > This would be optional in a machine built with dual-256x4 SRAMs.
>
> Isn't there already a fiting latch in the 18xx family ? (1883 ??)

The 1883 is a *7* bit latch with built-in decoders. We need no decoder
because the low 15 bits of the full address bus are going to each the
ROM and RAM sockets. There are no individual memory chips to select.
The 1883 is a great chip if you want to use several 6116's, for example.

> I would prefer 18xx chips prior to all others.

I'm willing to consider 18xx chips, but they are each several dollars,
while CMOS 4xxx chips are under a buck each, even in small quantities.
 
> But maybe space for two (or better four) 1855 MDU. Playing with such
> a beast is suposed to be a lot of fun. The same might be true for a
> 1879 RTC. These parts are still available.

I still have yet to hear why the 1855 Multiply/Divide Unit is worth
the real estate. Sure, it's a neat chip, but unless it has a purpose,
I can't see including it.

The 1879 Real Time Clock might be neat, but certainly an optional part.
 
> Also I would like to have an additional 1852 (i/o selectable via
> switch - but I only need output) to add an LCD display.

> There are some real cheap displays available with a paralell ASCII
> input - less than 25 USD for 16x2 characters.

I have several. BG Micro and other places sell used 20x4 and 16x2 modules
for $2 to $7. I must say that I hadn't considered that as an optional
peripheral. It might be worth laying out a 14 pin connector off of
the output port for that purpose.

> Let's get serious, the Elf99 is not suposed to be the real thing, used for
> real applications. It's just for the fun of doing it, and experimenting with
> unusual concepts (unusula from todays view). When ever you want to
> do real apps, any 6504+6532 or 68xx singlechiper will be more relevant.

Well... I wouldn't call a 6504/6532 board "relevant". Fun, probably, but
not current stuff. Fun this is, but if it isn't at least passingly
practical, nobody can do anything with it. I never really used my Elf
because I had no I/O beyond the Q LED and the switches. The main reason
for designing the Elf99 is to satisfy a 20-year-old desire to have a
"useful" Elf, and to capitalize on the present wave of nostalgia to
make a PCB cost-effective. I could always just bang out a single board
for myself with the iron-on transfers, but that's so much work that I'd
rather make something that I can share with a wider audience. It's risky
to consider putting up a few thousand dollars on a replica of a classic,
but if I maximize the appeal of the standard I/O ports, I'm likely to sell
a few dozen. If I can't sell 60, I can't afford to invest in 100 PCBs.
Received on Wed Nov 25 1998 - 15:07:10 GMT