Contact Us
  • Home
  • BLOG
  • MEGTRON6 vs MEGTRON7: When the Extra Loss Budget Is Worth It

MEGTRON6 vs MEGTRON7: When the Extra Loss Budget Is Worth It

MEGTRON6 and MEGTRON7 are the two Panasonic laminates that most 100G, 400G and 800G designs end up choosing between. They are the same family, they process the same way, and they are specified for the same reason: low loss at high frequency with FR-4-like manufacturability. The difference is one step in dissipation factor, roughly double the price, and a real question of whether the channel actually needs it.

The numbers

PropertyMEGTRON6MEGTRON7Why it matters
Dk (typical, 1 GHz)3.63.4Line width, impedance, propagation delay
Df (typical, 1 GHz)0.0020.0015Insertion loss per inch
Tg185–200 °C200 °C and aboveMultiple reflows, thick backplanes
CTE, Z axisAround 45 ppm/°CLower than M6PTH reliability in thick boards
ProcessFR-4-likeFR-4-like, tighter press windowYield, panel size
Relative costBaselineRoughly 1.5 to 2 times M6Board cost at volume

Datasheet values are measured at 1 GHz in most Panasonic documentation, and Df rises with frequency. At 10 GHz the gap between the two materials is still roughly a factor of two, and at 25 GHz both have moved up but the ratio holds. Always design against the curve at the operating frequency, not against the single number in the header of the datasheet.

What the Df difference buys in a real channel

Insertion loss in a PCB trace has two parts: dielectric loss, proportional to Df and frequency, and conductor loss, proportional to the square root of frequency and to copper roughness. At 28 GHz on a 4 mil line, dielectric loss is the larger of the two, which is why Df selection dominates at 112G PAM4.

High-speed stack-up
Layer build determines the channel loss

A rough rule for 112G PAM4 channels: on MEGTRON6, a 20 inch backplane channel at 28 GHz typically lands around 25 to 30 dB of insertion loss depending on stack-up and via count. On MEGTRON7 the same channel is typically 3 to 5 dB lower. Whether that matters depends on the SerDes: with a strong DSP and good equalisation, 28 dB can still close. With a marginal link budget or a long channel with several connectors, the 4 dB is the difference between passing and failing compliance.

Where MEGTRON6 is enough

100G and most 400G designs. At 25 to 26 GHz Nyquist, with channels under 15 inches and one or two connectors, M6 is usually sufficient and the equaliser has margin left.

Line cards and daughter cards with short reach. Much of the loss budget is spent in the connector and the cable, not the board. Spending on M7 for a 6 inch channel is money that does not show up in the eye diagram.

Mixed high-speed and power boards. M6 is more forgiving in lamination, and on thick power boards with many layers the press window matters more than the last 0.0005 in Df.

Where MEGTRON7 earns its price

Backplanes over 20 inches. Every inch costs loss, and the exponent works against you. This is the classic M7 application.

Orthogonal and direct-mate architectures. Long trace runs with multiple connectors and via stubs leave less margin for dielectric loss.

800G and 1.6T switch boards with 112G or 224G lanes. At 224G PAM4 the Nyquist frequency moves to around 56 GHz, dielectric loss dominates completely, and M7 or an ultra-low-loss material becomes the baseline rather than an upgrade.

Thermal margin. The higher Tg of M7 helps on thick boards that see multiple reflow cycles, especially when the board also carries heavy copper.

The copper matters as much as the laminate

Above 20 GHz, conductor loss from copper roughness can exceed dielectric loss on a well-chosen laminate. Moving from standard foil to HVLP or HVLP3 on M6 can recover as much loss as moving from M6 to M7, at a fraction of the cost. The two levers are not independent: an ultra-low-loss laminate paired with rough copper wastes most of what you paid for.

A practical sequence for a loss-limited design is: first reduce via stubs and layer transitions, then choose the copper foil, then choose the laminate. Most designs that jump straight to M7 while keeping standard foil and long stubs spend the budget in the wrong place.

Manufacturing differences that show up in yield

Press window. M7 has a narrower process window than M6. Thick backplanes with many prepreg layers need a press recipe developed for the material, not the shop's default high-Tg recipe. Ask for the press profile and the first-article cross-section before the production panel is released.

Drilling. Both drill cleanly, but M7's lower Df comes with a filler system that is harder on drill bits. Bit life is shorter, and hole wall quality should be checked on the first lot rather than assumed from the M6 experience.

Desmear and plating. Standard permanganate desmear works on both. For back-drilled vias, the depth control tolerance matters more than the material, because a stub that is 10 mil too long costs more loss than the Df difference between M6 and M7.

Availability and second source. M7 is a premium material with longer lead times and fewer qualified fabs. If the program needs a second source, that has to be planned at design stage, because a different laminate means recalculating every impedance and re-qualifying the stack.

How to decide

Model the channel first. Take the actual stack-up, the actual via count, the actual connectors, and compute insertion loss at Nyquist with both materials. Then look at the SerDes datasheet for the channel loss the part can equalise, and keep 3 to 5 dB of margin for manufacturing variation.

If M6 closes with margin, use M6 and spend the difference on better connectors, tighter stub control or improved stack-up symmetry. If M6 lands within 2 dB of the limit, M7 is the cheaper answer than a redesign after the first compliance failure.

The worst outcome is choosing M7 and then losing the gain to uncontrolled stubs, rough copper and long via transitions. Material selection is one lever among four, and it is usually not the cheapest one to pull first.

Need a quote or a DFM check on your files? pcba@superb-tech.com · WhatsApp +86 133 9608 1443