2
\$\begingroup\$

Input and output impedance is an important property of complex electronic components. Is it possible to physically measure them like we measure resistance?

Is it possible to measure it for BJT based transistor amplifiers and FET based transistor amplifiers? Also, can the same be done for complex ICs like FPGAs and microprocessors and PHY chips?

Related to this is the source of impedance in complex circuits, i.e what makes up the internal and external impedance. For high speed design we like to have matched impedance to prevent high speed effects in circuit operation. Usually this means making the signal track on PCB a 50 ohm trace and possibly use termination resistors also. A signal going into the input of the IC however has to travel on the PCB track, then through the pin on the package of the IC, through the bonding wire and then into pad and into the input buffer on the die inside the IC package. Do all these different components connected together have 50 ohm impedance each?

\$\endgroup\$
4
\$\begingroup\$

Is it possible to physically measure them like we measure resistance?

We can use a network analyzer to measure the reflection coefficient of a port of the component, and from this we can calculate the input impedance.

Is it possible to measure it for BJT based transistor amplifiers and FET based transistor amplifiers? Also, can the same be done for complex ICs like FPGAs and microprocessors and PHY chips?

Yes and yes.

A signal going into the input of the IC however has to travel on the PCB track, then through the pin on the package of the IC, through the bonding wire and then into pad and into the input buffer on the die inside the IC package. Do all these different components connected together have 50 ohm impedance each?

If the frequency is high enough that these features can not be treated as lumped elements, then yes, they should be designed as near as possible to 50 ohm characteristic impedance.

The traces in the IC package are routinely designed this way.

The ball of a BGA cannot practically be designed to have 50 ohm characteristic impedance, so we usually see some non-ideality (reflections) from the bga balls if the operating frequency is high enough.

The bondwires between the package and the chip itself also can't practically be designed to have 50 ohm characteristic impedance. Therefore a wirebonded package also will have some non-ideal behavior due to the bond wires. Flip-chip packages don't require bondwires so they can help to reduce the return loss.

\$\endgroup\$
  • \$\begingroup\$ wow, that is mind blowing. do I take this to mean that regardless of how precisely we design the PCB and use precise terminations, high speed effects will still occur regardless as the things inside the package are not all 50 ohm? \$\endgroup\$ – quantum231 May 5 '15 at 19:44
  • 1
    \$\begingroup\$ AFAIK, the pcb vendor can't really maintain the trace impedance better than +/- 5%, anyway. So if the package trace is exactly 50 ohms, and the on-chip termination is exactly 50 ohms, there could still be about 2.5% reflections. \$\endgroup\$ – The Photon May 5 '15 at 20:22
2
\$\begingroup\$

Typically the resistance of the bond wire and bond pad is minimal relative to 50 ohms so you don't usually have to worry about resistance of those components. You usually worry a lot more about inductance.

In a high-speed serializer IC often the output impedence is 50 Ohms and you terminate again at the receiver to reduce reflections.

Almost always the input stage of a high speed communications link is some kind of MOS buffer amplifier. The input impedance in that case is extremely high at DC and then the high-frequency input impedance is set by the input capacitance. For an FPGA or microprocessor most general-purpose IO is CMOS (so the "input buffer" would be a chain of inverters in the pad) which for a SERDES it may be a more "analog style" amplifier. In either case if you are interested in the high-speed input impedance use a Network Analyzer as The Photon suggests.

One last thing, at high-speeds engineers typically worry about S parameters, not lumped input impedance. So, a high-speed SERDES would most likely have the S11 of the input port in the datasheet, not an "input impedance" (i.e. so-and-so ohms vs. frequency).

\$\endgroup\$
  • \$\begingroup\$ hmmmm, So parameters. Better check it out, \$\endgroup\$ – quantum231 May 5 '15 at 19:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.