Smoked Crystal or PLL
Paul G
Posts: 10
I recently built a couple of stripboard based boards using the DIP prop. Everything was working until at the last minute decided to add a couple of smoothing caps to the power. I added a 100 MicroF cap on to the incoming power (before the regulators) and a 47 MicroF Tantalum cap between pins 9 & 12. I didn't have space to put the tant on the other side and 47microF was what I had to hand at the time. I had wired in VSS and VDD to both sides of the chip.
After powering on, the program didn't run. I changed _CLKMODE to RCFAST and it works, but trying to use the external crystal results in hanging, even without specifying the '+ PLLnnX' part.
I have removed both caps and replaced the crystal on one of the boards but it still hangs, so I assume that I have damaged the chip. I don't have a spare prop replacement for testing the moment.
I have searched the forum and found that the PLL circuitry is sensitive to incorrect polarities and spikes, but as far as I can tell, the tant was wired the correct way around.
My question is - what caused the problem and what can I do to make sure I don't smoke the next chip ?
Background - I am a software developer and learning the hardware side, so please go easy on a newbie.
Thanks
Paul
After powering on, the program didn't run. I changed _CLKMODE to RCFAST and it works, but trying to use the external crystal results in hanging, even without specifying the '+ PLLnnX' part.
I have removed both caps and replaced the crystal on one of the boards but it still hangs, so I assume that I have damaged the chip. I don't have a spare prop replacement for testing the moment.
I have searched the forum and found that the PLL circuitry is sensitive to incorrect polarities and spikes, but as far as I can tell, the tant was wired the correct way around.
My question is - what caused the problem and what can I do to make sure I don't smoke the next chip ?
Background - I am a software developer and learning the hardware side, so please go easy on a newbie.
Thanks
Paul
Comments
I would double check your crystal connections... A bad connection at the crystal could be the problem...
The Quartz crystals are generally more reliable, but it still can break under specific shock (such as slamming the board on table quickly and hard - it would always fracture a larger Quartz crystal - though the SMD crystals would ususally survive unscathed). Also, it's all too easy to pull the pins out and rip the wire off the crystal with specific brands (usually with the cheap crystals found in the cheaply-made RC toys).
Also, double-checking the crystal connection is best. AND, don't forget that the decoupling capacitors have to be of the right ranges as the PLL in the Propeller usually "resonate" rapidly - up to 300MHz as mentioned in few overclocking threads here - wrong capacitor choice, and the PLL's history. Low ESR capacitors are also extremely important - lossy higher-ESR capacitors would run out of juice quickly, causing the PLL clock glitches - another reason why they get cooked too easily. My opinions (to be taken with grains of salts), was that the PLLs were made with somewhat thin "wire" traces on the Propeller die, or could be the power input that's too "thin" for its current consumption ratings.
The PLL just happens to be (by testing) the weakest link on the chip. It doesn't take much noise on the power supply or ground to go from 3.3V to over 4V where damage will occur. The additional Vdd/Vss pins are provided for a reason ... to distribute the supply (and ground) currents across the chip so the voltage drops across the chip can be minimized at high currents. If they're not supplied or not properly bypassed, induced noise may have a few more tenths of a volt to ride on and the margins are tight.
Hence the reason why the SMPS have to be checked first and installed with the bulk capacitors (Supercapacitor is a big help, only if the power supply could handle the short-circuit of 2A for short time - 2 Farads), however, it also depends on the power supply designed and how well the board is grounded. I have used Propeller on a 3.3 volts line of a PC SMPS, and it worked fine - the actual mileages will vary.
Also, unrelated to power supply, I have wrote in my PDF about my own supercomputer, that the cleanness of clock signal is also EXTREMELY important, that is, if the clock beats is overlapped over few cycles, the PLLs will try to relock on what's essentially a dead clock pin (relative to the timing - couple hundred microseconds), the PLL, in some case, will short-circuit.
There are also few other faults toward the wimpy PLLs, which is also covered by few other threads, including what Sapheia mentioned.
All in fairness, it sure looked like it was already too late to incorporate the protection circuitry in the on-die PLLs...
Hopefully the Propeller II's PLLs will be able to handle all of the nasty faults better than its predecessor.
BTW, finally, few crystal oscillators are of somewhat low-failure parts, so I am not too worried (just don't pull the OE down for a specific times or bad things could happen....) I am using SiTime SiT8003 MEMS oscillator, though.
And... The wires. They're sometimes notorious for introducing few more millivolts into the chip which can be seen as a dip or spike in high-speed computerized oscilloscope... Did you put in the ceramic capacitor after the wires (disk or SMD wouldn't matter) on the Vss / Vdd? You will want to be SURE you don't introduce the microcontroller-killing pulses which can be induced by the wires... (PCB traces, no big deal. Dangling wires, big deal...)
What's amazing is that Parallax does not follow the rules, (number of capacitors and close proximity to the chip) and yet hundreds of Proto Boards, PEK Kits and PPDB don't seem to have a problem. Maybe no so much of a weal link.
[QUOTE=Hopefully the Propeller II's PLLs will be able to handle all of the nasty faults better than its predecessor. [/QUOTE]
Also, about bypassing the fried PLL, I don't think it's easy. Some PLL that's dying would often take the bypass trace with it, thus disconnecting the clock pins along with it, leaving the Propeller totally vegetable. Either you're lucky or you're not. But, you still can test to see if that "special" trace isn't touched (and "zorched") by the failing PLLs.
And... Fewer and very high-end microcontroller (PowerQUICC - I know it's a general purpose CPU intended for networking and embedded computing, but it can be used as microcontroller, anyways...) have so tight a voltage skew requirement with CPU die voltage of 0.95 V to 1.15V. Way too easy to get fried by the flaky power supply... (Although very well-designed power supply shouldn't be a problem, and it would be possible to run the MPC8377 on AA batteries for a while and MUCH longer on Lithium-ion battery, just the designing rules has to be followed...)
-Phil
However, PLL (Phase Loop-Locked) was the things that some peoples have been complaining about...
No wonder some of the clock chips are the highest failure parts... (The expectation, of course, would be the oscillator blisters [or pack, take your pick] as they already have the capacitor built-in. Yes, I have ripped one open out of curiousity. Two SMD high-Q capacitors are present as well... Not SMD ones, unfortunately. Regular Quartz and MEMS oscillators are typically fault-free as long as the right capacitors are supplied at the Vdd and Vss, usually an 0.1uF low-ESR SMD capacitor VERY close to the SMD oscillator or via the vias to the bottom.)
ALL - PLL's are Voltage sensitive as them use VCO - Voltage Controlled Oscillator.
In couple of the CPUs / MCUs datasheet, the PLL must be in the threshold of 0.1% voltage skew limitations, or you are going to be in trouble. Fried PLLs and/or messed up CPU core (ALU) frequency is the byproduct of what's going to happen if the rules are to be ignored. Not what you want seeing in your own hardware.
Very few oscillators wouldn't care as it got the on-die voltage regulator and DC charge pump. You gotta check the datasheet - they're tolerable within 10% voltage skew limitation, but not the CPU.
Oh yea, do not forget about the die temperature as it do and actually will affect the PLLs.
It is WHY all CPU's/FPGA's that have inbuilt PLL system NEED Very good decoupling.
And as Propeller use PLL's to --- NEED same rules as that others.
And Sapheia, few question I have been meaning to ask you... Is the capacitive value more important when choosing the decoupling capacitors than would be at the power supply rail? The reason I asked was because I am used to be putting in 0.1uF Low-ESR decoupling capacitors as usually recommended in the datasheet, while some people said 100nF is better (for some reason I disagree as its charges can be totally gone in a flash - up to 4 Amps in several nanosecond timing frames - the reason why I picked 0.1uF ones).
And, regarding high-quality Low-ESR capacitors (I usually use Japanese-made capacitor as I like to keep the chance of failure low), if trace width and length (on the power supply rail) should be carefully calculated to the nth degrees, if so - why?
Just use wide tracks for Vdd and Vss.
Yet, at 300mA, traces wouldn't have to be prohibitively thick, but it can and actually helps (it do reduce the chance of radiating nasty radio frequency around).
Leon and Sapheia... Good advices.
(P.S. Leon, you're right about the capacitive values... I was kinda asleep in my head - realized some part of my thread, now I felt like an idiot...)
So, Paul G, in the end, you may want to use very fast oscilloscope to see the spikes with a dummy device like NE555 timer IC on a simulator DIP-40 board to simulate the Propeller's load as you want it to be as close as the real thing so you could catch the nasty spikes and put in where it really need the decoupling capacitors. We do make mistakes, but sometimes it do help to make the same mistake, however in different ways, to find the solutions quickly.
As Leon pointed -- It is same value --- Only written diferently.
BUT that decoupling's capacitors NEED have very fast response time - As them are in place to SHORT all very fast transients.
To that You always have at last ONE 10-33uF(Varies on IC's current variations needs) Tantalum else other type of Capacitor That can HOLD and Smooth Voltage to IC
on all IC's that have Variable Voltage/Current consumption.
And what I knew and have been doing, the capacitors should be 0.1uF (some 1uF) and of very low ESR CLOSE to the processors (Propeller in this case), while you can place up to 2 inches apart a larger low-ESR capacitor at the power supply rail (my typical choice is Nichicon Low-ESR Aluminum Oxide Polymer capacitors). I have been using higher value at the entry point of power supply to help eliminate nasty ripples as well as the 60 hertzs DC pulses from shoddy wall adapter (it will still pulsate from the voltage regulator, not a good thing, nor pretty).
The only reason I use kinda larger value unless strictly listed in the datasheet, was because I have to assume the worst-case scenarios like using the Propeller board with lousy wallwart (the one that's heavy - with transformer in it) as 60Hz can often ruin your / our pleasure playing with Propeller, only to watch it getting turned on and off at 120Hz - which is why I use 1,200uF on mine.
I maybe not understand correctly. BUT Decoupling and BULK capacitors are not same You shall never optimize that by using only one capacitor.
If You use decoupling capacitor and BULK it need always be 2 capacitors in parallel. As them have different function.
To decoupling capacitors most used are Ceramic ones.
1uF are to small to be BULK and in most cases to BIG to be decoupling one.
Many times for correct function on Propeller in over-clocking as decoupling ones I use even 2 capacitors in Parallel - One 100nF and one 10nF. That are decided on Frequencies them need SHORT.
However, 1,200uF AlO3 Nichicon capacitor is to be used as a bulk capacitor AFTER the regulator(s) for the rest of circuitry while 0.1 to 1uF Ceramic low-ESR capacitors is to be used for decoupling the entry point of Vss / Vdd for the semiconductor ICs like Propeller, oscillator and finally a firmware ROM (EEPROM or Ferroelectric RAM), to isolate it from all of the nasty noise that could occur there. Also, trace length is apparently VERY important.
BULK Capacitor near REG are not same as BULK capacitor near Propeller. As You write Traces have some resistance and even if You have BIG Bulk near REG -- It will not prevent Voltage drops near Propeller.
And as I write in post before 1uF with 80MHz working Propeller are to small to be BULK and to BIG to be Decoupling one.
Usable values for Propeller are 10-100nF ceramic capacitors.
Delay between traces on SDRAM IO won't matter much, but on faster RAMs, like XDR and DDR1 - 4, it's much more important, as it becomes unacceptable at higher frequency, another reason for the trace matching.
I don't understand why it was working earlier, but I guess it was just on the edge of tolerance.
I told you I was a newbie.
Thanks for all the suggestions.
Paul
-Phil