This paper proposes a novel method for practical prediction of interconnect conductor surface roughness effect on multi-gigabit digital signals.


Abstract

The high-frequency content of state-of-the-art digital signals acts like a magnifying glass

for microscale PCB effects like glass weave and copper surface roughness. In this paper,

we present an experimental, numerical, and analytical investigation of the important role

that these effects play in defining the characteristic impedance of interconnects, the

associated attenuation and phase constant, as well as the appearance of resonances in the

insertion and return loss. By means of 3D EM simulations, macromodels are developed

and correlated to measurements up to 50 GHz using a test vehicle designed using

different prepregs, laminates and copper foils. As a result, recommendations and

guidelines are provided for proper material selection and trace routing that mitigate such

effects.

Summary: PCB laminates are inhomogeneous structures constructed from multiple

dielectric and metal layers. Thus, the anisotropic nature of PCBs introduces a variation of

the dielectric characteristics of the substrate with position and may also originate

resonances that considerably degrade the performance of interconnects. Despite this, it is

a common practice for design engineers to solely rely on vendor published ε and TanD

parameters obtained by a Bereskin and/or Stripline-resonator technique applied to smooth

resin cast of homogenous material. Similarly, the properties of metal layers are typically

characterized by bulk properties; at best, modified based on the model developed by

Hammerstad and Jensen to account for the increase on the attenuation due to copper

roughness. However, as communication speeds continue to rise, it becomes imperative to

analyze the impact of micro-scale effects not only on the attenuation constant, but also on

the phase constant and the characteristic impedance of the interconnects so that accurate

representations can be achieved for reliable circuit design.

In this paper, we study these micro-scale effects in detail. In particular, we analyze the

dependence of the electrical properties of interconnects with respect to their position on

the weave as well as the effects that result from the periodicity of the fiber glass bundles,

which can give rise to resonances. Furthermore, we discuss the mechanisms that produce

such resonances and propose an analytical equation for their prediction, validated by 3D

electromagnetic simulations and measurements. For this purpose, a 3D fabric weave

model based on actual prepreg layer measurements is created and simulated using an EM

simulator. A PCB test vehicle was fabricated with sets of traces running at various angles.

The predicted, simulated, and measured resonances are shown to be in excellent

agreement.


The focus of our investigation of copper roughness is not only on the additional

attenuation introduced by the roughness, but also on the additional time delay, which can

be directly extracted from our time-domain simulations. The investigation starts with the

common approach where the copper surface bumps are assumed to be periodic. A

parameterized model is developed which takes given surface statistics as input. The

statistical data to generate such models are taken directly from optical profile meter

measurements of copper foils. Bear in mind, however, that the large aspect ratio between

the roughness feature size and the overall interconnect length makes the use of these 3D

models currently unfeasible. For this reason, we propose a computationally inexpensive macromodel based on a frequency-dependent surface impedance model. After the models

are developed, we use them to predict the effects of copper roughness on signal

propagation and to characterize the copper foil by simulation. Our models are then

correlated to existing models and measurements.

For the experimental validation of the modeling results presented in this paper, we

determine the complex propagation constant (γ) and characteristic impedance (Zc) of

various high speed test lines. In our experiments we determine γ and Zc from line-line

measurements, which is desirable for a more accurate and systematic characterization of

the micro-scale effects on high-speed interconnects. Finally, as a result of our micro-scale

investigations of PCBs, we provide guidelines and recommendations for material

selection, stack-up optimization and trace routing not only for taking these effects into

consideration but also for mitigating them as much as possible.