64k, or even 128k, on a RAM A Board?!
by P. A. Forrester
There are probably many Nascom owners who would like to have a full
complement of RAM but who jib at paying over £100 for one of the 64k boards
which can be purchased, and so most of us soldier on with our old RAM A
boards. Some may have added another set of 8 4116s, using the ‘piggy-back’
method described in INMC-80 (1). However, now that 4164 dynamic RAMs can be
obtained for less than £5, I started to ponder whether a better solution might
not be to abandon the 4116s and add a block of 8 4164s at a cost of about £35.
After I had worked out how to do this, it seemed a shame to leave the other
block of 8 sockets unfilled, so I splashed out yet again and added another set
of 4164s to give me a total capacity of 128k. Since the Z80 can only address
64k at any given time, I had to find out how to page the additional memory in
and out of the address space. This note describes how all this can be
achieved. I will not attempt to give a detailed recipe of which tracks to cut
where, but rather, will try to give sufficient information so that anyone with
a bit of knowledge of digital logic should be able to follow what I have done.
The conversion to 64k can be done without the addition of extra support chips,
apart from a couple of inverters, and only requires simple mods to the
circuitry, but the use of two 64k blocks imposes the need for some means of
controlling which parts of the 128k of available RAM is coupled into the
memory-map at any given time and this requires some additional chips; the way
I have done it only takes 7 extra TTL parts costing in total under £2.
Let us start with a little background on dynamic RAM. This differs in
several respects from static RAM, such as the 6116 discussed in 80-BUS News
(2). Firstly, the chips we are interested in use address multiplexing;
secondly, they have only two data pins, an input and an output; and, thirdly
the data must be periodically refreshed. Also, because they are based on NMOS,
rather than CMOS, the 4116 consumes rather more power than the equivalent-capacity
6116. Let us briefly consider the implications of these statements.
Our understanding can be aided by Fig. 1, which gives the pin layout of the
4116 and 4164.
You see that the 4116 has only
7 address pins and the 4164 has 8.
To address 16 kilobits, we need 14
address lines, and so these have to
be fed to the chip in two parts.
First the lowest 7 lines of the
address bus are fed to the address
pins of the chip and these are
latched internally by applying a low
on the /RAS (row address select)
pin; then, the next 7 bits are fed
to the chip and these are latched by
pulling the /CAS (column address
select) pin low. It is one of the
functions of the board designer to
ensure that the addresses and their appropriate latching signals are fed to
the chip in the correct sequence. The 4164 operates in a similar manner but
requires that all sixteen address lines are coupled in two groups A0 – A7, and
A8 – A15. Setting up an address on the input latches gives access to one cell
so that we can read or write one bit. When R/W goes low, the bit presented at
the input pin D is latched provided /CAS is also low. The stored bit then