The Cheap Green Board That Quietly Eats Your Signal as the Frequency and the Heat Climb

Every builder reaches for it without thinking. The standard green circuit board, glass-reinforced epoxy laminate, is the default substrate of nearly everything, cheap, rigid, available everywhere, and perfectly fine at audio and low radio frequencies. Then someone tries to build a microwave filter or a clean transmission line on it and discovers, often the hard way, that the same material that behaved impeccably at 10 MHz is bleeding signal away at 2 GHz. The board has not changed. The physics of how its epoxy responds to a fast-changing electric field has simply caught up with the design. Dielectric loss in glass-epoxy laminate is the quiet tax that grows with frequency and with temperature, and the builder who does not budget for it watches power vanish into heat with no idea where it went.

The loss is not a defect in any one board. It is intrinsic to the material, written into the molecular structure of the cured epoxy, and it is strongly dependent on two variables the builder controls only partly: how fast the field oscillates and how hot the board runs. Understanding the dependence turns an invisible drain into a predictable quantity, one that can be calculated, designed around, or used to decide when the cheap green board must give way to something better.

What the loss tangent actually measures

The single number that captures dielectric loss is the loss tangent, written tan delta and sometimes called the dissipation factor. It comes from treating the dielectric's response to an electric field as having two parts, one in phase with the field and one lagging it. A perfect insulator stores energy and returns it all, its polarization following the field exactly. A real dielectric lags, and the lagging component dissipates energy as heat. The loss tangent is the ratio of the lossy, lagging part to the lossless, storing part, expressed through the complex permittivity:

tan(delta) = epsilon_imaginary / epsilon_real

where epsilon_real governs how much charge the material stores and epsilon_imaginary governs how much it dissipates. For a capacitor built from a lossy dielectric, the loss tangent is the ratio at any given frequency between the real and imaginary parts of the impedance, equivalently the ratio of the resistive to the reactive current.

For standard glass-epoxy laminate the loss tangent sits around 0.02 at 1 GHz, with the dielectric constant near 4.2 to 4.5. Compare that to a premium microwave laminate, where the loss tangent can be 0.002 or lower, a full order of magnitude better. That factor of ten is the entire reason microwave designers abandon the cheap board: every decibel of dielectric loss in a filter or a matching network comes straight off the signal, and a loss tangent ten times higher means roughly ten times the dielectric attenuation for the same geometry.

The polarization lag that grows with frequency

Why should the loss depend on frequency at all? The answer lies in how the epoxy's molecules respond to a reversing field. The cured resin contains polar groups, little molecular dipoles that try to align with the applied field. At low frequencies the field reverses slowly, the dipoles have ample time to swing into alignment and back, and they follow the field almost perfectly, dissipating little. As the frequency rises the field reverses faster, and the dipoles, hindered by the viscous drag of the surrounding molecular structure, can no longer keep up. They lag, and the lag is where energy is lost, converted to molecular motion, which is heat.

This is dielectric relaxation, and it gives the loss its characteristic frequency dependence. The published behavior of glass-epoxy laminate shows the loss tangent increasing rapidly up to about 100 kHz and then climbing more steadily up through the gigahertz range toward 100 GHz. The dielectric constant slides downward over the same span as the slower polarization mechanisms drop out one by one: from about 4.7 at 1 kHz to 4.5 at 1 MHz to 4.3 at 1 GHz. This gradual fall in permittivity with rising frequency is called dielectric dispersion, and it is the inseparable companion of the loss, the two linked because a material that dissipates must also disperse, a consequence of causality formalized in the Kramers-Kronig relations.

The dispersion matters beyond mere loss. A digital pulse is a superposition of many frequencies, and because each frequency travels at a slightly different speed when the permittivity varies with frequency, the pulse smears as it propagates. So the same molecular lag that dissipates a sine wave also distorts a fast edge, which is why high-speed digital designers care about the loss tangent of their boards just as much as microwave designers do.

A numerical estimate of the loss a transmission line suffers

Put the loss tangent to work and the abstract number becomes concrete decibels. The dielectric attenuation of a transmission line, the loss caused by the substrate alone, follows

alpha_d = (pi f sqrt(epsilon_r) * tan(delta)) / c (nepers per meter)

or in the more usable form for a TEM line, the dielectric loss in decibels per unit length scales directly with frequency, with the square root of the dielectric constant, and with the loss tangent. The published rule of thumb makes it tangible: a quarter-wavelength TEM transmission line on standard glass-epoxy laminate suffers about 1.8 dB of dielectric loss at 1 GHz. That is a startling figure. A simple quarter-wave matching stub, a component a designer drops in without a second thought at low frequencies, throws away nearly two decibels at 1 GHz, almost a third of the power, into the board.

Scale it with frequency to see the trend. Because the dielectric attenuation grows linearly with frequency for a fixed electrical length, doubling the frequency to 2 GHz roughly doubles the per-quarter-wave loss toward 3.6 dB, and at 5 GHz the same quarter-wave element bleeds away the majority of the signal. Run the comparison against premium laminate at tan delta of 0.002: the 1.8 dB becomes

1.8 dB * (0.002 / 0.02) = 0.18 dB

a tenfold reduction, turning a crippling loss into a negligible one. This single comparison, 1.8 dB against 0.18 dB for the identical structure, is the whole economic argument for premium substrate at microwave frequencies, and it is why the cheap board is fine at 144 MHz, marginal at 1 GHz, and usually unacceptable above 2 GHz for anything where loss matters.

Why heat makes a marginal board worse

Frequency is half the story; temperature is the other half, and it works in the same unfavorable direction. The molecular relaxation that causes the loss is a thermally activated process. Warming the epoxy changes the mobility of its polar groups and shifts the relaxation, and over the temperature range a board sees in service the loss tangent generally rises with temperature. The effect compounds dangerously with the loss itself, because dielectric loss dissipates power as heat inside the board, the heat raises the temperature, the higher temperature raises the loss tangent, and the higher loss tangent dissipates still more power.

The magnitude of the temperature dependence is large enough to measure cleanly. A study of glass-epoxy laminate across a wide temperature range found that cooling the material from room temperature down to a few kelvin reduced the loss tangent by about 70 percent and the real permittivity by about 9 percent. Read in the warming direction, that means heating the board substantially increases the loss tangent, the dielectric becoming markedly lossier as it warms. For a power application where the board already runs hot, this is a feedback the designer must respect: a transmission line carrying significant power on a warm glass-epoxy board loses more as it heats, and in an extreme case the loss and the heating reinforce each other toward thermal trouble.

The dielectric constant also drifts with temperature, which detunes any frequency-critical structure. A filter trimmed to its passband at room temperature shifts as the board warms, because the permittivity that set its electrical dimensions has changed. Combined with the large thermal expansion of the laminate, which physically changes the dimensions, a glass-epoxy microwave circuit can wander noticeably off its design frequency over a working temperature swing, a problem premium laminates with tighter thermal coefficients are built to avoid.

A numerical look at the temperature feedback

Walk the feedback through with numbers. Take a microstrip line on glass-epoxy laminate dissipating 5 watts of signal power, of which the dielectric loss claims, say, 20 percent, so 1 watt is deposited as heat in the substrate. If the board's thermal resistance to its surroundings is 15 degrees per watt, the temperature rise is

dT = 1 W * 15 degC/W = 15 degC

Now suppose the loss tangent rises by roughly 0.5 percent per degree near room temperature, a representative activation slope for epoxy relaxation. Over a 15 degree rise the loss tangent increases by

d(tan delta) = 15 * 0.005 = 0.075, i.e. 7.5 percent higher

so the dielectric loss climbs from 20 percent of the signal to about 21.5 percent, depositing slightly more heat, which raises the temperature a little further, which raises the loss again. The series converges because each step is smaller than the last, summing to a modest but real amplification: the final loss settles a few percent above where the cold calculation predicted. The feedback is gentle on a well-cooled board but turns vicious if cooling is poor, because a high thermal resistance multiplies the temperature rise and steepens every step of the loop. This is why a microwave power circuit on cheap laminate that tests fine cold can drift and degrade once it reaches operating temperature, the loss tangent climbing with the heat its own loss created.

The relaxation peak the loss tangent rides toward

The frequency dependence is not a featureless climb but the flank of a broad peak, and the Debye model of relaxation makes its shape explicit. A single relaxation process has a characteristic time tau, the time the dipoles need to respond, and the loss tangent it produces peaks at the frequency where the field oscillation matches that response time:

f_peak = 1 / (2 pi tau)

The loss factor itself follows the Debye form, in which the imaginary permittivity rises, peaks at f_peak, and falls again:

epsilon_imaginary(omega) = (epsilon_static - epsilon_infinity) omega tau / (1 + (omega * tau)^2)

where epsilon_static is the low-frequency permittivity and epsilon_infinity the high-frequency limit. Below the peak, where omega*tau is much less than one, the imaginary part grows linearly with frequency, which is the regime most ordinary radio work occupies, and it is why the loss tangent of glass-epoxy laminate climbs steadily through the megahertz and gigahertz range rather than jumping. The board's relaxation peak for its dominant epoxy process sits high, well up in the gigahertz-to-terahertz region, so across the entire useful radio spectrum the material is operating on the rising lower flank of its loss peak, the loss growing monotonically with frequency exactly as observed.

Temperature enters through tau, because the relaxation time shortens as the material warms following an activation law:

tau(T) = tau_0 exp(Ea / (k T))

where Ea is the activation energy of the dipole motion and k is Boltzmann's constant. Warming the board shortens tau, which slides f_peak upward and shifts the whole loss curve, and at a fixed working frequency on the rising flank that shift translates into a higher loss tangent. This is the molecular origin of the temperature dependence measured in the laboratory: the 70 percent drop in loss tangent on cooling toward a few kelvin is tau lengthening as the dipole motion freezes out, pushing the loss peak far above the measurement frequency and leaving the cold material nearly lossless. The model thus unifies both dependences, frequency and temperature, into one picture of dipoles racing a reversing field and losing the race by an amount that depends on how fast the field reverses and how warm the dipoles are.

Knowing when the green board has run out of road

The practical wisdom that falls out of all this is a set of thresholds. Below a few hundred megahertz, glass-epoxy laminate is excellent and the loss tangent is irrelevant for most purposes. Approaching 1 GHz the loss becomes a real budget item, and a designer must add the dielectric attenuation into the link budget rather than ignoring it. Above 1 to 2 GHz the cheap board becomes a liability for any loss-sensitive or frequency-critical circuit, and the order-of-magnitude better loss tangent of premium laminate justifies its cost. There is a further complication the loss tangent number alone hides: the glass-epoxy laminate's loss tangent is not even tightly controlled by manufacturers, varying from vendor to vendor and batch to batch, so a design that depends precisely on its value is building on sand. For anything critical, the realistic options are to measure the actual board, to design with generous margin, or to move to a controlled substrate.

The deeper lesson is that a substrate is not a passive backdrop to a circuit but an active participant that dissipates, disperses, and drifts. The cheap green board earns its ubiquity honestly at the frequencies and powers most work lives at, but it carries a tax that grows with every octave and every degree, a tax payable in lost signal and shifted tuning. The builder who knows the loss tangent, knows how it climbs with frequency and temperature, and knows the rough decibels it costs per wavelength can decide deliberately when to keep paying that tax and when the design has finally outgrown the board that started it.