I'm sure many of you analog engineers have heard the speech about the world being analog: Our ears, eyes, nose, mouth all work in analog so it doesn't matter whatever unnatural processing goes on in digital, because everything still starts and ends in an analog manner. And it's a speech I found I regularly had to deliver in the seventies and most of the eighties as academia churned mostly digital engineers out of the educational processors.

Well, that's not the speech I'm going to give you. I don't disagree with the line taken and to a first order simplicity that view of the world is true. But it doesn't go anything like far enough in showing that digital doesn't really even exist at all as a separate discipline! It is only a minor, special case of analog.

Saturated Amplifiers

When you look at a digital data stream it is either a number of parallel paths or a multiplexed serial stream. The existence or non-existence of a signal at a particular clock point determines a '1' or an '0'. The clock itself, of course, is analog. Data are, to me, just modulation streams in their simplest form. They are directly comparable to Morse Code with the presence or otherwise of a carrier for two different duty cycles. Passing the signals through logic gates is merely the process of either saturating or turning off an amplifier.

The amplifier form is analog, powered by analog. The voltages switched are analog. Intel has now shown us in their multi-level flash that you can, of course, have mid-level signals, more analog levels, to store much more in the same devices. This process was first used by ISD in their multi-level analog storage ICs for audio. What limit is there in such a system? If you can resolve signals to X mV and can read the signal out of the noise floor it is probable that 10-12 levels would be resolvable. The voltages going down on logic parts might be counter-productive at some point in limiting memory.

As noted, clocks are surely analog, as is clock distribution; to a less obvious first glance so are the interfaces and I/Os: All are transmitter/receiver styled and without analog techniques for re-clocking, latching, etc. they wouldn't operate.

The Digital RF

When we hear of developments in processor speed and the revolution (or is it a reverse-revolution, counter-revolution?) of going to copper in the fabrication, we have now witnessed demonstrations of CPUs operating at 1 GHz. And that is incredible, there is no denying. But what will be the throughput speed of products using such processors?

I witnessed a group of very professional PC board manufacturing engineers discussing, just a couple of months ago, the problems they were having with maintaining impedance across their boards. The results their OEM clients were getting did not tally with the theory they were using. They were ignoring, totally, the imaginary part of the impedance!

It really has to be wake-up time in board houses, OEMs, PC makers, chip makers. This is RF you guys are playing with! Taking the motherboard that gave you 40-MHz throughput with your 70-MHz processor will still give you 40-MHz throughput at 200 MHz, and at 1 GHz as well.

Now if this is not the realm of the true analog engineer then I don't know what is. As an RF engineer myself (although I am frequency-challenged to 40 GHz) I think of my caste as being the epitome of analog engineering with the lower frequencies being highly-rewarding in different but lesser ways! But with "digital" engineering climbing up to real frequencies it will be interesting to watch the manufacturers meet the real challenges.

Culture Differences

There is little to compare the vast chasm that exists between the way that a digital engineer and an RF engineer think about things, and how they behave and work. The culture gap is so large that I have seen RF engineers forced out of jobs by their surroundings and fellow engineers after extremely short tenures. And if you are in any doubt about where the jobs-for-life are today in our industry try and recruit a group of RF engineers. If you really want to be hard on yourself, try and recruit them with relocation required.

So far, the best motherboards that I have seen tested for throughput have come from DEC; they have been vastly superior to other US product and dramatically superior to some of the product coming from Asia. With the swallowing of DEC by Compaq and some subsequent deals about Alpha technology to Japan I hope that group is not split up. But even the DEC motherboards could not throughput the CPU speed. And everything is working against the designer: multi-layers, multiple power lines, multiple frequencies, poor connectors, demands for rock-bottom cost. But can it be done? Of course, but the companies need to stop thinking digital and start thinking analog.

Outside Solutions

Powering PCs has become extremely complex with varying voltage levels in a manner that even the analog IC industry gasped over as the specifications were released, and repeatedly changed. But outside, analog, design-power has come through with operating devices allowing the CPU manufacturers to run their chips with lower power consumption, at higher speeds and with excellent monitoring and protection.

Monitoring and control are going to get even more sophisticated, and things are not going to get any less complex as the voltages continue to go down (to reduce the squared-component of power) and the currents climb towards, and probably over, 20 A. The busbar needed to carry this kind of current around is an unknown for the PCB manufacturers. It is time they and the CPU manufacturers found some outside help. If you are an RF engineer with layout experience to the GHz region and have also done some multi-layer board work there is a new career for you, just waiting.

The future for engineering is still analog.

By: Paul McGoldrick
Sr. Technology Editor, EDTN

Analog Main | Product of the Week | Columns | Editorial | Tech Notes

		Click here to get your listing up.