|
|
IBIS-AMI是最近的热门话题
本帖最后由 stupid 于 2010-5-18 14:09 编辑 " g% P0 r1 y+ h( ^0 j# D
: v7 W6 k" T* J( ^# B( tYou cannot predict the performance of a high speed serial link without having an accurate model for the transmitters (TX), the channel, and the receivers (RX). These days, the channel model, ultimately represented by its S-parameters, is the easy part.
) k, J! o3 @; }; ^' V
$ a* j: C" m- O# | ^8 lIf you can describe a single differential channel in terms of the S-parameters from one end to the other, you can incorporate it into most simulators. This could be a touchstone file with an .s4p extension, for example, signifying a 4-port S-parameter data set.
' r# S0 p4 P, y6 j% }8 `4 a. J* ^9 `7 v, S, J
Even better, if you can include the S-parameters for one or two adjacent channels, you can include the impact from uncorrelated channel to channel cross talk. This requires a .s8p or even a .s12p touchstone file.+ s+ \& _& k7 B: K1 s
1 J F2 C, R7 z1 ^
But what about the TX and RX devices? IBIS models have been used to describe the TX and RX properties of devices which includes the switching rise time, output impedance, input gate capacitance, and even a simple RLC package model. However, in virtually all high speed link transceivers, equalization and clock recovery circuits are an integral part of the I/O circuitry. Up until now, there were no hooks in the IBIS model to include these features. ! X" G7 ]; T' ]/ Q" ~
6 z$ G4 g$ ~3 O![]() / D5 j$ I% a8 j4 Y" k
Extending IBIS models to include these “analog” features is the purpose of the IBIS-AMI(Algorithmic Modeling Interface) spec. The paper by Todd Westerhoff and his co-authors, presented at DesignCon 2010, “Predicting BER with IBIS-AMI: experiences correlating SerDes simulations and measurement,” provides a concise introduction to the features of IBIS-AMI models and some of the learning curve SiSoft and IBM went through to bring their simulation tools into excellent agreement./ a3 @+ w+ ~: z! L% z& j9 Q: \/ R
' n% `# I9 [" d; l
This paper introduces the role of the AMI spec and alternatives to circuit simulation for link analysis. Under the IBIS-AMI umbrella, there are two parts to a link model: the passive channel, from the pads on the TX chip to pads on the RX chip and the serdes Tx/Rx analog characteristics.
O$ U! m8 n$ U2 j- g2 T0 ?' G5 e3 H9 ]3 w4 e9 g* ?0 G. B3 Q2 Q/ J
The impulse response of the channel, which can be obtained from the SDD21 response of the channel, can be used to predict the statistical properties of the eye directly, or with a convolution integral, calculate the transient waveforms at the receiver. No circuit simulation need be performed, just signal processing.5 N# v S) R% O
1 z: t( |/ r4 X$ ^5 {
The second part of the AMI model is the algorithm to process the received signal. This can be how the impulse response is transformed by the equalizer into a new impulse response or how the transient waveform is processed by the equalizer. Even the behavior of how the clock recovery circuitry acts on the received transient waveform can be described in the algorithm. : c% A: j# `3 f: `" j" y1 y1 @
The rest of this paper describes the correlation between IBIS-AMI models simulated in IBM’s internal simulation environment, which has been extensively verified with measurements, and the Quantum Channel Designer, the SiSoft link simulator environment. As seen in the figure above, at the end of their program, the agreement between the two simulation environments turned out to be very good. 7 S- x3 R, L, @) n7 t3 f) z3 d z
S x( c" r4 m! l( Y
If you would like a good introduction to IBIS-AMI models, this paper is a good starting place. |
|
/1 
发表于 2010-5-10 22:03
QQ好友和群
QQ空间
腾讯微博
腾讯朋友
微信
收藏
支持！
反对！
楼主
发表于 2010-6-30 12:30
