O
Member
【Editorial Discussion】
A stack-up entered as a single, isotropic dielectric constant can look fine on a simple delay coupon and then disagree once you simulate a coupled line, a different orientation, or a second coupon on the same panel. It is tempting to “fix Dk” until one result matches. Before you do that, it is worth writing down whether the model is missing anisotropy, or whether the glass-weave position under that particular line is the larger term.
1) Isotropic Dk is a convenience, not a complete material. Many laminates have a different permittivity through the thickness than in the plane of the cloth. A microstrip delay coupon is often dominated by the through-thickness field; a tightly coupled pair, an edge-coupled filter, or a line with a strong horizontal field can see more of the in-plane value. If the coupon delay matches and the coupled-line even- or odd-mode result does not, an anisotropic stack is a reasonable next model — but only after you have used the same frequency-dependent dielectric data in both, not a single constant Dk fitted to one trace.
2) Glass-weave position is a different error. The local glass-to-resin ratio under a narrow line, or under one member of a pair, can change delay even when the datasheet Dk is unchanged. Two coupons that were supposed to be identical, or a pair that shows unexpected skew, may be telling you about weave location rather than about a missing tensor in the material card. If rotating the routing, moving the line, or comparing several coupons on the same panel moves the result, treat weave position as a first suspect, not as a rounding error.
3) Do not fold both problems into one fitted number. Raising or lowering a single isotropic Dk to force a coupled-line match will usually break the coupon, or the next structure, or the next frequency. A cleaner split is: keep a causal, frequency-dependent dielectric; add anisotropy only if the same stack still disagrees across structures that see different field orientations; and treat coupon-to-coupon scatter that tracks routing location as a weave or process term, not as a new Dk.
4) Write down which comparison you actually made. Coupon delay versus a single line in the same orientation is not the same test as coupon versus a coupled pair, or versus a line rotated relative to the weave. Without that note, a later “the material model is wrong” discussion cannot be settled.
Open question: the last time an isotropic Dk made a coupled-line or delay result disagree with the coupon, what actually closed the gap — an anisotropic stack, glass-weave position, or a datasheet Dk that was never the right starting point? Please share the kind of structure and how you proved it; no customer names or unverified numbers needed.
A stack-up entered as a single, isotropic dielectric constant can look fine on a simple delay coupon and then disagree once you simulate a coupled line, a different orientation, or a second coupon on the same panel. It is tempting to “fix Dk” until one result matches. Before you do that, it is worth writing down whether the model is missing anisotropy, or whether the glass-weave position under that particular line is the larger term.
1) Isotropic Dk is a convenience, not a complete material. Many laminates have a different permittivity through the thickness than in the plane of the cloth. A microstrip delay coupon is often dominated by the through-thickness field; a tightly coupled pair, an edge-coupled filter, or a line with a strong horizontal field can see more of the in-plane value. If the coupon delay matches and the coupled-line even- or odd-mode result does not, an anisotropic stack is a reasonable next model — but only after you have used the same frequency-dependent dielectric data in both, not a single constant Dk fitted to one trace.
2) Glass-weave position is a different error. The local glass-to-resin ratio under a narrow line, or under one member of a pair, can change delay even when the datasheet Dk is unchanged. Two coupons that were supposed to be identical, or a pair that shows unexpected skew, may be telling you about weave location rather than about a missing tensor in the material card. If rotating the routing, moving the line, or comparing several coupons on the same panel moves the result, treat weave position as a first suspect, not as a rounding error.
3) Do not fold both problems into one fitted number. Raising or lowering a single isotropic Dk to force a coupled-line match will usually break the coupon, or the next structure, or the next frequency. A cleaner split is: keep a causal, frequency-dependent dielectric; add anisotropy only if the same stack still disagrees across structures that see different field orientations; and treat coupon-to-coupon scatter that tracks routing location as a weave or process term, not as a new Dk.
4) Write down which comparison you actually made. Coupon delay versus a single line in the same orientation is not the same test as coupon versus a coupled pair, or versus a line rotated relative to the weave. Without that note, a later “the material model is wrong” discussion cannot be settled.
Open question: the last time an isotropic Dk made a coupled-line or delay result disagree with the coupon, what actually closed the gap — an anisotropic stack, glass-weave position, or a datasheet Dk that was never the right starting point? Please share the kind of structure and how you proved it; no customer names or unverified numbers needed.
