I like that you mention the wavelength as this only works up to a certain frequency. As you go high in frequency, your component pad will have to go down in width. Essentially, you'd want the pad to be thinner than the trace. In our radar stuff, I just don't put an 0402 cap as that's too big for what the frequency allows. Very good explanation.
If I understand you correctly, your chain of capacitors should all be within a fraction of your wavelength? If you start pushing higher percentages of a wave length should you split up your capacitors by chunks having large impedance traces between sections of capacitors? Is their an EMI impact if youre doing this alot? Probably one of those things you have to build and find out.
Good video. At long long last somebody did it, cuz it's getting boring ti explain about caps in TL. It will be good to hear about TL to PAD impedance mismatch (for RF and HS dig): REf. plane cutouts, impedance transformers and teardrops when pads are LARGE ( some RF filters have PADS 3x3mm and with line 0.3). I mean real calculation.
Very nice video Zach! I've been following your videos for some time now and they really helped me understand some things. I think it would be really nice to show us a practical example of designing a PCB -> perhaps a small sample of 4 layer PCB - how would you start, how would you decide how to use stackup, where would you place vias (under capacitors or next to), maybe comment on return paths, copper pour in GND and PWR planes... I think that would really sum up some of your recent videos! :)) I'm also sure a lot of guys would appreciate it!
Hello. Thanks for a start! It would be very valuable for me to understand the choice of the reference layer, and termination (pullup-down). For example, HDMI(TMDS)-type interfaces are better based on +3 rather than GND.
great video! it's interesting that the thickness of the pads become relevant above 10 GHz. Is there a general rule for figuring out which what the upper limit for component size on a transmission line should be based on operating frequency? I imagine this is a loaded question as the parasitic resistance and inductance of any caps will change with size as well!
"I do not know of a formula that is used to select an upper limit on component size for a given frequency. Component manufacturers will design their packages to have sufficiently low package parasitics so that they can accommodate the rated frequency range at the specifications given in the datasheet. This is why something like RF transceivers will have maybe QFN or similar SMD packages, you're basically eliminating pins and vias as well as all the parasitics that come with those packages/IOs. In general smaller SMD packages will be better at higher frequencies, but this might make routing at high frequencies more difficult depending on your trace geometry needed in your stackup for your target impedance. Sometimes you have to route in with a taper. The component manufacturer will (or should) provide some data that allows you to qualify performance in the relevant frequency range so that you can judge whether your link will meet your performance specs. Just as an example, we recently used an HMC802 attenuator in a design for a switchable radio link operating at high power, the datasheet lists a reference insertion loss that you would expect along the link based on the manufacturer's eval board design. Your job as the board designer is to make sure the rest of the interconnect is designed properly to make sure the board parasitics don't create some new problem."
I should put an asterisk next to that statement because it does depend on the package size and the dielectric constant of the capacitor's dielectric. The capacitor could have very high dielectric constant, so even at low RF frequencies, a large case capacitor on a signal line could still look like an impedance discontinuity and the signal would interact with this capacitive impedance. You also have to remember that at low frequencies, the input impedance approaches infinity (as we would expect as we near DC), so be careful not to overgeneralize.
Hi Zach, just one question here. We deeply understand the size of components inserted to a transmission line highly determine input impedance that signal sees. That's also why designer puts AC coupling cap near driver side like TX in PCIe topology. There's one question here, in that way why IC houses still recommend PCB designer to add void underneath AC cap in an adjacent return path (e.g. solid ground plane) when transition vias are needed? It looks like there will be a dip on input impedance when AC coupling cap is inserted, so we need to make a discontinuity area on purpose to compensate this dip. Based on the ADS simulation from some articles, this will compensate roughly 5-7 ohms up for PCIe gen3 as signal travels through AC cap.
The only reason one can claim is that it would reduce a small amount of parasitic capacitance around the pads and input traces coming into the capacitor, but this would only matter with very small capacitors. This is one of those rules where some people say they never remove the ground and they never have a problem, and other people will always remove the ground and they never have a problem. Like you have mentioned, it is important to simulate it in your specific stackup to see if you actually need to recover some of the impedance.
8:50 - How small does a capacitor needs to be so that it’s small enough to be considered zero for the hyperbolic tangent function of “Lc” plugin? Like 0805 package? How about those regular through hole electrolytical capacitors? Thanks.
It depends on the frequency in relation to the propagation constant in the body of the capacitor. This is kind of a difficult problem to analyze so I do not want to tell you a specific package size.
Hey Zach, I am really enjoying your lessons on signal integrity. I have a question about the input impedance formula you are using. As far as I am aware, the input impedance is not a constant value but is actually a function of time which transitions between values as the signal propagates and interacts with different components. If so, what exactly does that formula represent and how does it characterize the overall impedance of a transmission line and a load.
I believe what you're referring to is the surge impedance, which is the time-domain analogue of input impednace. I'm an RF and optics guy so I'm always used to frequency domain explanations. As I recall surge impedance is just an impulse response to the input waveform. This is one of my favorite types of questions because impedances are defined as being functions of frequency. However, from the perspective of transfer functions, an impedance can be viewed as a function that transforms current into voltage and vice versa. Thus, an impedance function (which we would call a Z-parameter in network analysis) also has an impulse response function that is the inverse Fourier transform of the impedance vs. frequency function. The voltage response would be convolution of the impedance impulse response with the input current waveform.
@@brokenicry thanks! I suppose the output impedance of an IC is also the input impedance of the transmission line, it's just what label it's given in the equation?
Yes, it is the driver's output that is matched to an input impedance. In other words, if you're looking outward from the driver toward's the load, you would ideally want the driver's output to be matched to the input you see looking at the output pin. The point is that everything downstream from the load will affect what the input impedance looks like.
An Eternal Perspective 7 "Then will [matter] be [destroyed], or not?" The Savior said, "Every nature, every form, every creature exists in and with each other, but they'll dissolve again into their own roots, because the nature of matter dissolves into its nature alone. Anyone who has ears to hear should hear!" Peter said to him, "Since you've explained everything to us, tell us one more thing. What's the sin of the world?" The Savior said, "Sin doesn't exist, but you're the ones who make sin when you act in accordance with the nature of adultery, which is called 'sin.' That's why the Good came among you, up to the things of every nature in order to restore it within its root." Then he continued and said, "That's why you get sick and die, because [you love 8 what tricks you. Anyone who] can understand should understand! "Matter [gave birth to] a passion that has no image because it comes from what's contrary to nature. Then confusion arises in the whole body. That's why I told you to be content at heart. If you're discontented, find contentment in the presence of the various images of nature. Anyone who has ears to hear should hear!" www.gospels.net/mary ruclips.net/p/PLxxvBg1S_QqK5bzP7R57d187fZCbtS0Yx
Even a capacitor or resistor is not infinitely small, it can experience wave propagation. In fact the higher the dielectric constant of the dielectric, the smaller the wavelength inside that dielectric, thus the more likely that capacitive medium is to see wave propagation at lower frequencies.
Thanks for answering my question. A good explanation !
I'm learning a lot from these videos .
Awesome! We’re so glad 🙂
I like that you mention the wavelength as this only works up to a certain frequency.
As you go high in frequency, your component pad will have to go down in width. Essentially, you'd want the pad to be thinner than the trace. In our radar stuff, I just don't put an 0402 cap as that's too big for what the frequency allows.
Very good explanation.
If I understand you correctly, your chain of capacitors should all be within a fraction of your wavelength? If you start pushing higher percentages of a wave length should you split up your capacitors by chunks having large impedance traces between sections of capacitors? Is their an EMI impact if youre doing this alot? Probably one of those things you have to build and find out.
Good video. At long long last somebody did it, cuz it's getting boring ti explain about caps in TL.
It will be good to hear about TL to PAD impedance mismatch (for RF and HS dig): REf. plane cutouts, impedance transformers and teardrops when pads are LARGE ( some RF filters have PADS 3x3mm and with line 0.3). I mean real calculation.
Thanks for sharing these valuable information!
Glad it was helpful!
Very nice video Zach! I've been following your videos for some time now and they really helped me understand some things. I think it would be really nice to show us a practical example of designing a PCB -> perhaps a small sample of 4 layer PCB - how would you start, how would you decide how to use stackup, where would you place vias (under capacitors or next to), maybe comment on return paths, copper pour in GND and PWR planes... I think that would really sum up some of your recent videos! :)) I'm also sure a lot of guys would appreciate it!
You got it, I'll put together a little example so that we can see some good practices.
Hello. Thanks for a start!
It would be very valuable for me to understand the choice of the reference layer, and termination (pullup-down). For example, HDMI(TMDS)-type interfaces are better based on +3 rather than GND.
great video! it's interesting that the thickness of the pads become relevant above 10 GHz. Is there a general rule for figuring out which what the upper limit for component size on a transmission line should be based on operating frequency? I imagine this is a loaded question as the parasitic resistance and inductance of any caps will change with size as well!
"I do not know of a formula that is used to select an upper limit on component size for a given frequency. Component manufacturers will design their packages to have sufficiently low package parasitics so that they can accommodate the rated frequency range at the specifications given in the datasheet. This is why something like RF transceivers will have maybe QFN or similar SMD packages, you're basically eliminating pins and vias as well as all the parasitics that come with those packages/IOs. In general smaller SMD packages will be better at higher frequencies, but this might make routing at high frequencies more difficult depending on your trace geometry needed in your stackup for your target impedance. Sometimes you have to route in with a taper.
The component manufacturer will (or should) provide some data that allows you to qualify performance in the relevant frequency range so that you can judge whether your link will meet your performance specs. Just as an example, we recently used an HMC802 attenuator in a design for a switchable radio link operating at high power, the datasheet lists a reference insertion loss that you would expect along the link based on the manufacturer's eval board design. Your job as the board designer is to make sure the rest of the interconnect is designed properly to make sure the board parasitics don't create some new problem."
Why is it that a small capacitor (smaller than wavelength of signal or rise time/ fall time of signal) does not affect the impedance much?
I should put an asterisk next to that statement because it does depend on the package size and the dielectric constant of the capacitor's dielectric. The capacitor could have very high dielectric constant, so even at low RF frequencies, a large case capacitor on a signal line could still look like an impedance discontinuity and the signal would interact with this capacitive impedance. You also have to remember that at low frequencies, the input impedance approaches infinity (as we would expect as we near DC), so be careful not to overgeneralize.
Hi Zach, just one question here.
We deeply understand the size of components inserted to a transmission line highly determine input impedance that signal sees.
That's also why designer puts AC coupling cap near driver side like TX in PCIe topology.
There's one question here, in that way why IC houses still recommend PCB designer to add void underneath AC cap in an adjacent return path (e.g. solid ground plane) when transition vias are needed?
It looks like there will be a dip on input impedance when AC coupling cap is inserted, so we need to make a discontinuity area on purpose to compensate this dip. Based on the ADS simulation from some articles, this will compensate roughly 5-7 ohms up for PCIe gen3 as signal travels through AC cap.
The only reason one can claim is that it would reduce a small amount of parasitic capacitance around the pads and input traces coming into the capacitor, but this would only matter with very small capacitors. This is one of those rules where some people say they never remove the ground and they never have a problem, and other people will always remove the ground and they never have a problem. Like you have mentioned, it is important to simulate it in your specific stackup to see if you actually need to recover some of the impedance.
@@Zachariah-Peterson Thanks Zach.
Thank you! That was informative.
Glad you enjoyed it!
8:50 - How small does a capacitor needs to be so that it’s small enough to be considered zero for the hyperbolic tangent function of “Lc” plugin? Like 0805 package? How about those regular through hole electrolytical capacitors? Thanks.
It depends on the frequency in relation to the propagation constant in the body of the capacitor. This is kind of a difficult problem to analyze so I do not want to tell you a specific package size.
I would like to see a real live example with realistic numbers plugged into the equations.
Hey Zach, I am really enjoying your lessons on signal integrity. I have a question about the input impedance formula you are using.
As far as I am aware, the input impedance is not a constant value but is actually a function of time which transitions between values as the signal propagates and interacts with different components. If so, what exactly does that formula represent and how does it characterize the overall impedance of a transmission line and a load.
I believe what you're referring to is the surge impedance, which is the time-domain analogue of input impednace. I'm an RF and optics guy so I'm always used to frequency domain explanations. As I recall surge impedance is just an impulse response to the input waveform.
This is one of my favorite types of questions because impedances are defined as being functions of frequency. However, from the perspective of transfer functions, an impedance can be viewed as a function that transforms current into voltage and vice versa. Thus, an impedance function (which we would call a Z-parameter in network analysis) also has an impulse response function that is the inverse Fourier transform of the impedance vs. frequency function. The voltage response would be convolution of the impedance impulse response with the input current waveform.
Thank you for great explaination....
Glad you liked it
I'm confused; shouldn't the driver IC's output impedance be what is being matched to the transmission line and load?
You're right... It's just the way he's explaining it that may confuse some. You wanna match the driver to the line to the receiver input.
@@brokenicry thanks! I suppose the output impedance of an IC is also the input impedance of the transmission line, it's just what label it's given in the equation?
Yes, it is the driver's output that is matched to an input impedance. In other words, if you're looking outward from the driver toward's the load, you would ideally want the driver's output to be matched to the input you see looking at the output pin. The point is that everything downstream from the load will affect what the input impedance looks like.
An Eternal Perspective
7 "Then will [matter] be [destroyed], or not?" The Savior said, "Every nature, every form, every creature exists in and with each other, but they'll dissolve again into their own roots, because the nature of matter dissolves into its nature alone. Anyone who has ears to hear should hear!" Peter said to him, "Since you've explained everything to us, tell us one more thing. What's the sin of the world?" The Savior said, "Sin doesn't exist, but you're the ones who make sin when you act in accordance with the nature of adultery, which is called 'sin.' That's why the Good came among you, up to the things of every nature in order to restore it within its root." Then he continued and said, "That's why you get sick and die, because [you love 8 what tricks you. Anyone who] can understand should understand! "Matter [gave birth to] a passion that has no image because it comes from what's contrary to nature. Then confusion arises in the whole body. That's why I told you to be content at heart. If you're discontented, find contentment in the presence of the various images of nature. Anyone who has ears to hear should hear!"
www.gospels.net/mary
ruclips.net/p/PLxxvBg1S_QqK5bzP7R57d187fZCbtS0Yx
Nice!👍
Thanks! 👍
you need a Smith chart!
Ugggh... I hate Smith charts!
How about some practical examples on how to measure these impedances?
I just recently got a little Libre VNA that is rated up to 6 GHz, we'll put something together ASAP!
Wait, what about lumped elements (capacitance will have its own impedance regardless of size)? There is some mistake in your reasoning!
Even a capacitor or resistor is not infinitely small, it can experience wave propagation. In fact the higher the dielectric constant of the dielectric, the smaller the wavelength inside that dielectric, thus the more likely that capacitive medium is to see wave propagation at lower frequencies.