Tech Ideas Page
Some technical/scientific ideas relevant enough to share,
some of which I might quite well implement as soon as I
have some things, including but certainly not limited to funding
arranged. I guess I have the copyright on them anyway, and you'll
need serious skills to implement them yourself, so I see no
big problem putting on the net, if you're interested drop me
an email or sign the guestbook.
Where to get those up to date digital/electronics parts?
Want to start a new company? Try (re)selling electronical parts,
either by mail or as a store. Some of the one's I could find that
actually have some of the hottest latest DSP/memory/cpld/fpga/pc
materials have a profit margin of roughly 800% (my own personal
guess...), and than they still don't have all the parts you might
want as a serious freetime or professional engineer. Also, a
lot of magazines, even interesting one's (I'm talking with a
fairly large dutch bias here) seem to remain stuck at designs
with materials that could be called obsolete or at least not
very innovative if taken for what they are, such as (small) pic's
and microcontrollers that make even a straightforward Z80 card
look interesting in comparison.
I recently found out that both the Siemens 80c535 and the familiar
80(c)51 uc's are based on the 8080 processor, and I just can't
resist to utter my dismay over this. You can buy cmos Z80 chips
(even the peripherials) that run at ca. 40MHz (tried it myself!)
for about $3 a piece, and there are loads of compilers and
interpreters for it around. So it looks nice all these
microcontrollers and their fancy PC based programmers, but
technologywise and price-wise it compares unfavorably with
a fairly simple Z80 card.
Agreed, you don't get a fancy package
with many IO pins, you'll have to buy your own 74AC573 buffers
for each 8 bits, but at least they'll run at 180Mhz, and cost
less together than a uc chip... Throw in a nice cmos memory ($4 for
64 K ?) with a simple batery backup, and maybe a fast cache
memory (20nS, 32kx8, 1.5mA rest current, for lets say another $4)
and we're back in the 80's but this time fast and cheap.
would be nice to throw in a few fast multipliers, some good alu's,
or is that asking too much? Another simple idea is to take a
68c000 costing about $12 add some of the same cheap peripherals,
use an 'old' atari 1024 or so with turbo C as a devkit, and
suddenly you can have serious programs on a fairly small and
cheap card for very little money also.
I guess the point is that
real innovations are not so much found here. Its nice to have
clock rates from 0 to 30 MHz in 0nS, and that everything is
nicely low power, and a lot faster, but seriously innovative
new digital designs are not to be found this way. An that pretty
much holds the same for 'building' your own PC, the range of
possible motherboards, processors and memory chips does not show
that much real variation. Come to think of it, what's wrong with
breadboarding a pentium (except for te still relatively high
price, the cheapest I could find was a $50 90MHz pentium)?
I did breadboard designs that were quite more complex then the
roughly 500 connections needed to connect up one (say 6 or 7
640 pin boards with half the holes filled), not even counting
the superfluence of a lot of the pins for simple ideas. Sure beats
a pic or a lot of random logic, and a microcontroller.
Just have a pentium, a simple startup program in either battery
backed sram or (e)eprom, a cristal or RC net, some IO buffers
(or if you must a pio or
something),and you can program your own microcontroller in wathever
language you choose from your PC (I would have to check if a
relocating linker is needed, but fundamentally I see no problem).
Seems worth a relatively cheap printed circuit board design to
me, and I don't think a lot of glue is needed to even fit on
a small completely cheap set of dynamic rams to give it some more
workspace at the expense of more supply power.
Some small, not too ambitious design ideas
When checking for readily available digital computing parts,
checking everthing from ttl adder ladders, and mos multiplyers to
the latest logic families, my main observations are (more later)
that the 'ttl' type series, including the 74 HC(T)'s are not extended
with interesting logic blocks for doing alu operations.
One may take the latest Cisc, risc and dsp CPU's as their followers,
which in itself makes sense, but there is no standard do it
yourself way of dealing with them that I am aware of, and
many of these products have heavy price tags.
For instance the TMS320C80 is targeted to cost $50 in 2000 at
reasonable quantities, but if you try and order one now you'll have
to draw roughly $1400 from your credit card or checkbook. Not
nice to try the maximum solder time on, but if you want 2Gflops
it could be worth it.
On the low end, I recently found that the TMS320c32, a tiny memory,
multiply and ALU op per 2 clock cycles (20-60MHz) dsp with
some of the now standard DSP facils like 2 dual ported on chip DMA
controlers, internal parallellism up to 6 parts, 32+8 bit
operations, support for zero overhead short loops and control,
on board cache, and supposedly not to hard to glue into a
system (will check data-sheets soon), is 'handed out' by TI
in single item quantities for $10,= ... That indeed renders a
lot of random logic designs obsolete, or too expensive.
It seems to be a bit of exception, because another component out of
their dsp series (I think I checked the fixed point 320c50)
costs $50 from some web post-based company, and these parts are
not to be obtained from the handful residential electronic part
resellers I tried.
I peeked at the Sharc prices in the same way, and found out that the
$29 in couple of ten quantity advertised by Analog Devices is not
excactly matched by the current resellers practice, it cost several
hundred dollars, in fact even more than their devkit which includes
a codec on a pc-connectable board for $179.
Try to buy your own (Xilinx) cpld's or fpga. It did find various
manufacturers (incuding Philips) that make them, and luckily put
datasheets on the web, but prices on the market are 10 times the
advertised prices or so.
A mechanical tinted idea: the breadboarding robot
A breadboard implementation of a digital or electronic design
has a lot of advantages. First: it can be built fairly fast,
second: it is very much correctable, third: it has good
visual feedback on some of HF properties, fourth: it is
easily extenable and testable.
A differential geometry ameanable computer design
bezier subdiv in fpga, preferable with someprocessing nodes
such as pentium II's or alpha's, few ti asics for communication
struct,set of sharcs or 320c32 (or 80'sif you have the money),
use some AC buffers if needed and some 12nS static mems (64 wide)
This is just general, the special props could be: