Basics for copper wired chip to chip communication

[PShah]

I have a very weird assignment/project. The link http://www.cns.cornell.edu/documents/JeffKashIBMTJWatsonResearchCenter.pdf" has an eye diagram in pg4 of the slide. That eye diagram has been constructed using copper wired interconnect for chip to chip communication. Now, I have no idea what is meant by chip to chip communication nor an example of it has been given to me. I have been asked to simulate such a communication so that I get the eye diagram. The Question is how do I do it because I don't have any knowledge about the system,communication technique,signal etc. How do I assume all this?Kindly help me in how to begin and what it really is all about. Will be really relieved if some examples,circuit etc are provided to get an idea of what is in store for me. Thank you

 

[yungman]

I have just very little experience in optical communication so I just joint in only. Don't take my words too seriously.

First question is: Have you study this kind of stuff before? I don't think this is an easy assignment at all depend what level are you in.

Eye pattern is formed by random 1 and 0 and if you just display on the scope. It is the standard way of looking at the rise and fall time, look at how distinct the bit level is. If the rise time is too slow, you never reach the required level before the next bit change. The first one without equalization, you can see the eye closed as it is all smeared. The one with equalization is much cleaner because you get better rise time on all the edge. That's all it means.

There is no info given about the equalization mechanism.

About chip to chip. My understanding is just means the wire is bonded direct to one chip on one end and to the other chip on the other end. This can reduce most interface compare to one that goes from chip to bonding pads, then to connector before the wire. Each interface create reflection and introduce ringing onto the signal and fuzzy up the eye pattern.

I don't really know where to start on this because it sounds like you supposed to know these to do the simulation So I am just going to guess! Do you know how to draw an eye diagram and why it look like this? I have problem with my scanner that I am getting help in the computer forum here, or else, I can step by step show you how to draw the eye pattern so you can think of how to do the simulation.

If you are asked to do the copper link that connect "chip to chip". Then all you have to do is to simulate the transmission line of the coax. You have to put in the conduction loss( surface impedance of the wire and the shield) the loss tangent of the dielectrics of the coax line of the given length.

Since it is chip to chip, you also need to know the output impedance of the transmitting chip and the input impedance of the receiving chip. With that, you can calculate the phasor( point to point amplitude of the forward and reflecting wave) in any location along the coax and then use the data to plot the eye pattern. This is my understanding.

 

[PShah]

@yungman: Thank you for the detailed response, its quite clear now regarding the eye diagram except for the simulation part of the transmission line of the coax. How do you simulate an analog copper wired chip to chip circuit? What will be the elements and the source and receiver of the signal, the kind of signal. How do I create a chip in Matlabs simulink? An illustration with an example for such a kind of communication would really help since I am totally new in circuit simulation of this kind.

 

[yungman]

I don't know Matlabs, I only know Microwave Office and I don't think that will even plot time domain plot.

I don't think it is easy because the input and output impedance vary with frequency and random 1 and 0 can be wide band. You need to ask your professor what is the assumption of the input and output impedance. I know when I was in the field, the transceivers have high input impedance and low output impedance so we actually put a 50 ohm in series between the output of the transmitter chip and the input of the coax. Then on the receiving end, we use a 50 ohm termination. We then use a 50 ohm coax line as the copper wire. With this assumption, then the first approximation is the input and output impedance at the two ends are 50 ohm and does not have any reflection, you get rid of the worst part of the simulation. With that, your problem become very simple. All you need is to plot the attenuation of the coax line with the group delay. If you have to take into consideration of the varying input/output impedance, you are going to have fun ( not!) modelling that. You have to translate the S parameters at a broad frequency range and do simulation one frequency at a time......NOT fun and you have my sympathy!

You need to ask your professor how he model the input and output of the chip first. Again, don't treat me as an expert, I only serve as a sounding board for you as I have very limited experience on fiber communication. I am more into RF circuits only.

 

[PShah]

Thank you once again. I have a very basic(and maybe a naive question), can RF circuits be used for this purpose. This is basically an analog simulation with special requirements. So what are required for communicating over wired line between two chips. U have mentioned about impedance, apart from that are there any special signal which is to be considered and any modulation scheme?Thanking you for your time and patience.

 

[yungman]

What you have is RF circuits. I am just not familiar with your transceivers and what kind of modulation scheme. Usually in transmitter and receivers, the input and output impedance are very frequency dependent and is usually given by the S-parameters at different frequencies. That's how we do matching to get max power transfer. But in your case, you might have a broad band signal and I don't know exactly how to deal with it in this case. That is the reason you need to clarify with your instructor what kind of modulation you are dealing with and what is the bandwidth of the signal. Also the input and output impedance are the transceivers.

The kind of modulation will tell you what frequency components you are dealing with. For narrower band, it is a lot easier to match, but if the modulation involve wide band, then it is going to be hard. So talk to your instructor before going any further.

 

[yungman]

I was kind of curious this morning and snoop around on OC192 and 10G logic and I found this data sheet:

http://www.digchip.com/datasheets/parts/datasheet/340/KGL4201.php

You might luck out! According to this data sheet, the input and output are internally 50ohm terminated. There is no max clock period which implies that it can support from DC to 10G signals. I am not going to really look into this, but if one is 50ohm, I suspect others are.

If this is true for other digital logic used in OC192 etc., then your simulation is just the coax line and it's attenuation and dispersion. Verify with your professor.

But also, you need to find out what dielectric material your professor expected you to use because you need to get the data of the loss tangent. Also, ask him can you assume the conductor is copper that the conductance $\sigma_{Cu}= 5.8\times 10^7\;$ . You need that to calculate skin depth which in turn the resistance of the wire at 10GHz.

 

[PShah]

He did not tell me anything,he told to read literature and find out how to the communication. I really cannot follow any of the building blocks and components which you mentioned. Thank you in trying so hard to provide the resources and links. U mentioned that I need to focus on the communication link, so what should be the components?

 

[yungman]

He did not tell me anything,he told to read literature and find out how to the communication. I really cannot follow any of the building blocks and components which you mentioned. Thank you in trying so hard to provide the resources and links. U mentioned that I need to focus on the communication link, so what should be the components?

If he did not give you anything to go by, how are you supposed to do a simulation. These are all the important parameters for simulation! From my little finding, you can assume the source and termination impedance are both 50 ohm, and all you need is the copper wire which in your case has to be a coax for single end. You want a 50 ohm coax.

The next step is to have the dielectric material so you can calculate the loss and specify the conductor material to calculate the resistance. Both of these are needed to model the transmission line. With that then you can work on what is the wave form at the output.

I must be missing something, this to me is a very difficult simulation. In my RF field, I don't even do time domain simulation like what you need to to produce the eye pattern. I hope someone else can join in and take a look at what I said so far. As I said, I am not an expert, only like half a year in the OC48 and 192 job.

What is your background and what is your major? What year are you in?

 

[PShah]

I am in a fix! The link I provided is the only information looking at which I am supposed to do a software simulation. I am in my final year Masters in Computer Science and supposed to Phd in Electrical and Computer Science. You have mentioned about dielectric constants etc, des this not apply to hardware level? How do I simulate in software a wired transmission line with the necessary components?

 

[yungman]

I am in a fix! The link I provided is the only information looking at which I am supposed to do a software simulation. I am in my final year Masters in Computer Science and supposed to Phd in Electrical and Computer Science. You have mentioned about dielectric constants etc, des this not apply to hardware level? How do I simulate in software a wired transmission line with the necessary components?

What you are trying to simulate is hardware. What is to simulate software? You show the hardware that is copper wired chip to chip, it is hardware, it is RF! The only thing I can think of is simulating the transmission line. If you are going to major in Electrical engineering, you should have studied electromagnetics already. Here is the transmission line model I refer to. What you have is a Lossy Transmission line with series resistance due to copper wire and skin effect. You have conductance shunt loss due to dielectric loss. Here is 3 articles talking about the lossy transmission model:

http://en.wikipedia.org/wiki/Transmission_line

http://www.ece.uci.edu/docs/hspice/hspice_2001_2-269.html

http://w2.cadence.com/appnotes/TransmissionLineApplicationsInPSpice.pdf


If you don't have any instruction, this is the only thing I can think of. Or else, what is to simulate? As I said, you need to talk to the professor. You are guessing and so am I! So far, I am giving you the tools, I have no idea how to simulate in time domain. I don't know PSPICE, maybe you might have better luck with that.

 

[PShah]

Thanx for all the support and resources. The professor's view point is the system in the pdf which I posted must be simulated. Now, the pdf belongs to Agilent Technologies and IBM whoc have their own ADS software platform. So, if they have simulated using ADS then it may be possible to do a software simulation using MATLAB' tollbox which as all these features. Anyway, you are right I need to work on this. Thank you once again.

 

[yungman]

When you find out what to do, can you update us so I can see the result?