best balance of performance and board cost

Since the turn of the 21st century, there has been intensive research toward the development of embedded optical channels for transporting high-speed digital signals within printed circuit boards. I suspect that an alternative to copper traces might be commercially viable by 2035 or sooner, depending on the development of semiconductor devices with integrated photonics to transmit and receive signals through those channels.
Finding the low-loss material that will provide the best balance of performance and board cost for a given application is more complicated than simply comparing laminate data sheets and prices.
The development of embedded optical waveguides and integrated photonics is driven largely by the capacity demands (as well as the power consumption) of high-performance routers and network switches, whose backplanes may span more than 20 inches. Signal attenuation resulting from dielectric and conductor losses is a major concern for designers of those backplanes, as are signal reflections caused by impedance variations resulting from shifts in the dielectric constant of laminates with frequency. Signal propagation delay, which is governed mostly by laminate dielectric constant, and trace crosstalk are also spurring the development of optical channels.
Optical channels would be immune from noise, virtually loss-less, and electrically independent of surrounding material. They would be the conduit for high-speed signals, while copper traces elsewhere in the boards would comprise the remainder of circuits. However, creating those channels involves changes throughout the entire infrastructure of electronics manufacturing, including semiconductor materials and fabrication processes, IC and PCB design tools, and IC packaging technology, beyond the development of optical materials compatible with high-volume, panel-based PCB manufacturing processes. That will take time.
Meanwhile, PCB materials with stable Df values on the order of 0.003 up to at least 10 GHz are necessary to meet channel loss budgets in such current high-speed digital applications as network line cards for 40-Gbit/s and faster data rates. Various materials, some of them widely used in RF applications, have Df values low enough to satisfy the loss budgets of high-speed signal paths on 40-Gbit/s Ethernet line cards, for example, within safe margin. These materials cost more than regular FR-4 laminates, so hybrid stackups are common that dedicate high-speed nets to low-loss layers and less-critical circuits to layers of FR-4 for economy.
 A Suitable Low-loss Material
Finding the low-loss material that will provide the best balance of performance and board cost for a given application is more complicated than simply comparing laminate data sheets and prices. Data sheets do not reveal which materials involve relatively more or unusual processing steps during PCB fabrication, which can raise manufacturing cost. Consider Rogers 4350B and Panasonic Megtron 6, which have similar low Df and Dk values, have been used extensively in RF applications, and are increasingly being used for high-speed digital products. Both are based on hydrocarbon resins; the Rogers resin has a ceramic filler. Neither laminate is available clad with quarter-ounce copper. The thinnest foil available for the Rogers material is half-ounce, and for Megtron 6, one-third-ounce. Both materials are available with low-profile foils to prevent signal reflections at high frequencies. The Rogers core material is essentially perfectly flat and repeatable, aiding impedance control; the Panasonic material slightly less so. The Rogers material is at least twice as expensive as Megtron 6.
Rogers offers three prepreg choices for bonding the 4350B laminates: a 4-mil prepreg that is available in two glass styles and one that is 3.6-mils thick with one glass style. Rogers discourages etchback of the material, advises against using a single layer of prepreg in high-layer-count, single-lamination stackups,and recommends cap construction. Manufacturers have to adjust the lamination cycle for fabricating boards when the Rogers material is involved because of the restriction on using a single layer of prepreg. The Rogers prepregs for the 4350B cores require higher pressure for proper lamination than do the Panasonic prepregs,which process no differently than conventional FR-4 materials.
Eight laminate thicknesses are available. Megtron 6 laminates come in 18 thicknesses, complemented by a wide range of prepreg thicknesses and glass styles, including various tightly woven, so-called flat-glass styles to avoid impedance variation caused by fiber-weave effect. Resin evenly coats the surface of those tight weaves. Three different percentages of resin content can be selected for several of the Megtron 6 prepreg glass styles. The most significant contrast betweem the Rogers material and the Panasonic material is that Megtron 6 laminates the same as conventional FR-4 materials; no incompatible pressures, temperature, movement, or cure time are involved.
What is the upshot of the differences between Rogers 4350B and Panasonic Megtron 6, beyond their raw material costs, considering that their electrical properties are alike? The most significant contrast is that Megtron 6 laminates the same as conventional FR-4 materials; no incompatible pressures, temperature, movement, or cure time are involved. Hybrid boards can be built in a single lamination with inner layers of relatively inexpensive FR-4 materials and an outer layer or layers of Megtron 6, using foil construction or cap construction. Moreover, the wider selection of Megtron 6 core and prepreg thicknesses and resin content eases stackup development and impedance control.
Many PCBs have been built using Rogers 4350B material for very high-speed digital circuits. It is a proven choice from a functional perspective, and a highly predictable material from a manufacturing perspective, with well-established fabrication protocols. Though its use is routine, it is somewhat more complicated to process than Megtron 6. The fact that fabrication complexity and yield have an inverse relationship is worth recognizing at the outset of design. Manufacturing yield may not be a concern if you only need a few boards, but that is certainly not the case for volume production.
from:Sierra Circuits