The community of IC designers is apparently split into two groups: those who do current-mode circuits and those who don't do them.

But what about the integrated circuits they design? Are they split into two classes as well? After all, we find assertions of the kind "current mode is better than voltage mode" in many papers.

In this talk, I will defend a seemingly paradoxical position. I will argue that there is no current mode that can be divided from a voltage mode. I will argue that assertions like "current mode is faster than voltage mode" are incorrect. I will argue that the current-mode-versus-voltage-mode divide is unreal like a dream.

But I will also argue that those who dreamed that dream have found very good circuits and techniques that should by all means be re-integrated into main-stream analog IC design. In this sense the current mode is very real.

Some among you may know that I did my doctoral thesis on current-mode video-frequency filters. During my doctoral studies, I read lots of papers on current-mode circuits, and found lots of assertions like "current-mode circuits are faster, less noisy, more linear than voltage-mode circuits," which sometimes came with references to other papers. Naturally I wondered why this would be so.

So I started to trace references back as far as possible, until I had a tree containing close to 100 papers --- but no evidence at all. All I found were assertions like "this current-mode circuit can be expected to have these performance advantages over this class of voltage-mode circuits." Which is quite different from "current mode is better than voltage mode."

I also tried to find out how current mode can be defined in a precise way such that it divides clearly between current-mode circuits and voltage-mode circuits. Some say that current mode is "current input signal, current output signal." This is, however, a system-level definition and is totally unrelated to performance questions. After all, any voltage-mode circuit can be converted into a current-mode circuit with one current-to-voltage and one voltage-to-current converter. I will come back to this topic soon.

Let's try another definition then: "Signals are represented by currents only." --- Well, even a current mirror has signal-dependent voltages and might be seen as a voltage-mode circuit under this definition. Next try: "Signals are represented by currents as a linear function." --- Then we have defined the differential pair as a current-mode circuit.

I could go on with this for hours, and in the end you would have a long story that tells you what I mean when I use the term "current-mode." But you would still not have a precise definition. And I think this is good so. A story explains a lot although it cannot clearly divide, but a definition only divides and explains nothing. I like explanations better than divides because explanations are useful for me.

And this is why I write in the paper that a clear definition would destroy the explanatory power of the current mode.

Let's go back to the "current input current output" definition of current mode. This one is as precise as it gets, but it explains nothing at all. Let me show you.

To find out whether the signal mode makes a performance difference, one needs to compare two circuits where only the mode differs but all other things are equal. Mahattanakul and Toumazou attempted to do this, and found that voltage-mode filters are slightly better. But is this so?

What they did was to analyse these two Gm-C filters in great detail and to compare them with a figure of merit. Their analysis is very well done, and I don't doubt that it is correct, but unfortunately there is a difference between these circuits other than their signal mode.

Both filters have the same feedback loop. In the voltage-mode filter, the signal is fed in with an OTA. In the current-mode filter, it is fed out with an OTA. So here the noise of the additional OTA is filtered, and here it is not.

But you cannot load this output, nor can you load this input.

In order to use the filter, the voltage-mode filter needs an output voltage buffer and the current-mode filter needs an input current buffer. This time, the noise of the current buffer is filtered and the noise of the voltage buffer is not filtered. The slight advantage of the voltage-mode filters that Mahattanakul and Toumazou found then disappears.

And you see something else: The centre of the circuits is identical. Any possible performance difference can only be caused by the circuits that feed the signals into and out of this centre. So any possible performance difference can only come from the transistor-level design of these subcircuits and not from the signal mode!

So let's focus on transistor circuits. I will now briefly compare a commercial voltage opamp connected as a unity-gain voltage buffer to a second-generation current conveyor used as a unity-gain current buffer.

I built the latter myself; it is a video-frequency circuit. In order to make the two buffers comparable, I scaled the frequency axes of the voltage-buffer curves by a factor of fifty.

On the top, you can see the transfer function of the current buffer (a) and the voltage buffer (b). They are very similar. This (c) curve is the open-loop transfer function of the opamp.

The bottom graph shows you the lower of the terminal impedances. This (a) is the input impedance of the current buffer, and this (b) is the output impedance of the voltage buffer. The latter is much lower because the voltage buffer is bipolar and the current buffer is CMOS.

Now I want you to look at what happens whit the output buffer impedance because of feedback. At low frequencies it rises with 20 dB per decade; above the opamp unity-gain frequency it is constant. But around the unity-gain frequency, some overpeaking occurs. The feedback only reduces the output resistance for frequencies below one fifth of the unity-gain frequency. --- A similar argument can be made for the effect of feedback on the reduction of harmonic distortion. You find a more detailed discussion including a sensitivity analysis in the paper.

With what I have said until now, I can now explain why current-mode circuits are often faster than voltage-mode circuits.

Current-mode circuits often have less feedback than voltage-mode circuits, so they are not subject to overpeaking. Then current-mode circuits often are less complex than the voltage-mode circuits they are compared to, meaning that they have fewer internal nodes in the signal path. So they are in fact faster.

But this is not a technical necessity. Do you remember Bram Nauta's work? He built very fast voltage-mode filters with OTAs that have no internal nodes at all! Or do you know the work of Arbel, Mucha, Palmisano, Palumbo, Pennisi, and so on? They built current-mode opamps which are built to use stronger feedback and have higher complexity, and have a performance very similar to the performance of standard voltage-mode opamps.

What have we got until now? I told you that I never saw evidence for a performance advantage coming from the mode of a circuit. I told you why I think there cannot be such an advantage. And I even told you that there are no precise definitions of the term "Current mode." --- So, does the current mode exist? --- My answer is no. --- If you look at "current mode" itself, you will find nothing. As such it is empty and insignificant.

This is actually the same for all engineering terms. Whenever you want to find out what something really is, you will not find anything that has real existence by itself.

What you will find are lots of dependencies with other terms, persons, and organisations; dependencies which can very well be explained but cannot be defined. They can very well be identified, but they also change all the time. Current-mode only makes sense if it is understood as a dynamical net of dependencies. But in this conventional way, the current mode is very real.

Just by the way, the philosophy that things only exist dependently but are empty of any inherent existence is not new. It is in fact very old. It is the basic philosophical principle of Buddhism. But there the principle extends to everything, ever to the "self" of a person. As John Donne said, "No man is an island, entire of itself." So the answer to the question "Who am I" will also be a long story about dependencies and not a simple definition. And if somebody gets attached to a definition instead of taking a wider view, that person will create suffering for himself and others. Or, going back to the problem at hand, people who cling to the term current-mode as such will have problems and cause problems.

I wrote an invited paper about this philosophical side of the problem in the "Australian Journal on Information Systems;" you'll find it in the references and also on my web site. In that paper, I wrote something about these dependencies. Let me give you a very brief summary of how I think new terms are made and develop.

A researcher tackling a new problem is always overwhelmed by its complexity. He must decide at which aspect he looks first, what he plans to consider later, and what he neglects. This means he reduces the scope of his thinking on purpose. As a result, he will start to see patterns, like Barrie Gilbert who saw that voltages are incidental in translinear loops. Maybe he will even coin a new term, like Gilbert who coined the term "current-mode."

This starting point may trigger lots of research, and in the beginning, people must defend themselves by saying "we're doing something new and great." That's alright, after all, one needs a lot of time until one is ready to defend a new idea. In this first phase of research, a bit of propaganda is not only allowed, it is essential for progress.

But one should not forget that it all started with a few persons who chose a restricted view to get a grip on a complex problem. This restriction remains if it is not overcome consciously. After a while, it stops driving research and instead starts blocking researchers. All that remains visible then is the propaganda, which cannot be excused anymore, which is even very bad for the reputation of current-mode research and makes many researchers distrust the knowledge made by the current-mode community.

And what do I conclude from all this?

The term "crrent mode" neither divides nor qualifies circuits. It has served us well in the past by giving researchers a direction and, with a bit of legitimate propaganda, by protecting researchers from too early critique.

I believe that the time where current-mode research has to be protected is over. Strong ideas, concepts, and circuits have come from it, and they should now be re-integrated into main-stream analog IC design.

The propaganda part should be stopped immediately. Too many young researchers believe the propaganda and consequently direct their research in a direction that does not promise good results anymore.

Finally, and more philosophically, beware if you find anything which seems to have quality all by itself. Because then you have just not found out through which dependencies that quality arises.

Thank you for your attention!