glen@proconel.com 
In the latest datasheet for the ATMega48PB Rev E the Full swing crystal oscillator option has been removed from the clock fuse settings. I believe it has also been removed from the PA device as well. Does this mean that the chip can no longer operate at 20MHz using a crystal? The datasheet only shows the low power crystal oscillator going up to 16MHz. We currently use about 30K a year of these devices using a 20MHz crystal.
Wow. I agree with you.
For the ATmega48PB:
The data sheet says the Int Osc using an external Xtal max'es out at 16 MHz, in spec.
When using an External Clock fed into Xtal1 the max is 20 MHz.
I was going to check the spec's on the M48PA to see if it might be an option, but the Atmel site has locked up twice trying to download the data sheet.
Michael or anyone else at Atmel listening in? (To both issues)
JC
Edit:
Welcome to the Forum.
In the latest datasheet for the ATMega48PB Rev E the Full swing crystal oscillator option has been removed from the clock fuse settings. I believe it has also been removed from the PA device as well. Does this mean that the chip can no longer operate at 20MHz using a crystal? The datasheet only shows the low power crystal oscillator going up to 16MHz. We currently use about 30K a year of these devices using a 20MHz crystal.
Does sound ominous.
Did you test 20MHz with the low power setting ?
Most Xtal oscillators do not have a brick-wall sudden failure, it is more a gain transfer effect, which means if you carefully select a Lowest-ESR Crystal, with lowest CL, you can nudge more MHz out of any Oscillator amplifier.
You can vary Vcc to check Oscillator Start margin, as well as add a stepped series resistor to the crystal.
Of course, smaller SMD crystals, are trending to increase ESR, which does not help..
theusch 
I believe it has also been removed from the PA device as well.
Why would you think that?
In the latest datasheet for the ATMega48PB Rev E the Full swing crystal oscillator option has been removed from the clock fuse settings...
Indeed, apparently a change in this datasheet rev.
Strange--I dropped a '168PB into an existing Mega88 design, and we always set the fuses for "full swing", and it operated OK...
Does this mean that the chip can no longer operate at 20MHz using a crystal?
Some discussion on that for the '328PB in this thread:
https://www.avrfreaks.net/forum/m...
and here, about post #20:
https://www.avrfreaks.net/forum/a...
Not yet reflected in the migration app note:
http://www.atmel.com/images/atme...
=====================
At your volumes, I'd get Atmel involved and keep us posted as is being done for the '328PB.
glen@proconel.com The Mega48PA datasheet still shows the Full swing crystal oscillator option, however I have just received a PCN notifying a die change which removes this option.
....
and here, about post #20:
https://www.avrfreaks.net/forum/a...
...
At your volumes, I'd get Atmel involved and keep us posted as is being done for the '328PB.
Good links, I see this is also date-code related - Post #27 says
[" I received my production parts - date code 1543. That's between the dates of cases that I reported to Atmel (Full Swing Oscillator not working and O3B/OC4B not working). If they hadn't already fixed the issues it would have been too late, I suspect.
My initial test shows that the low power oscillator is much better behaved. I tested it with 8MHz, 14.7MHz, 16MHz and 20MHz crystals and it appeared to work fine (on the bench) in each case. It wouldn't run faster than 12MHz before, same crystals, same load capacitors.
Just for the fun of it, I tried the Full Swing Oscillator and it seemed to work fine with 8MHz, 14.7MHz and 16MHz crystals but not the 20MHz crystals. That is an improvement over the Engineering Samples which I couldn't get to run faster than 12MHz with the Full Swing Oscillator."]
The Mega48PA datasheet still shows the Full swing crystal oscillator option, however I have just received a PCN notifying a die change which removes this option.
It's all unclear if 20MHz & Xtal is now in 'no mans land' ?
A Quick filter at Digikey on 20MHz shows CL from 7pF to 50pF and ESR from 10 Ohms to 200 Ohms. Those are big ranges.
10 Ohm ESR parts come in HC49/US for something of a premium, but they could be useful for margin testing.
There are ones that seem ok with low CL, moderate ESR and SMD packages ?
| NX3225SA | MHz Crystal | 20MHz | ±15ppm | ±15ppm | 8pF | 50 Ohm | Fundamental | -10°C ~ 75°C | - | Surface Mount | 4-SMD, No Lead (DFN, LCC) | 0.126" L x 0.098" W (3.20mm x 2.50mm) |
Do I dare to make a comment about how little energy Atmel put into documentation :(
theusch ... I have just received a PCN notifying a die change which removes this option....
I can't find that on Atmel site. Can you post the guts?
WC154601?
http://media.digikey.com/pdf/PCNs/Atmel/WC154601.pdf
I feel like I'm watching an old friend slowly waste away...
|
"Experience is what enables you to recognise a mistake the second time you make it." "Good judgement comes from experience. Experience comes from bad judgement." "Wisdom is always wont to arrive late, and to be a little approximate on first possession." "When you hear hoofbeats, think horses, not unicorns." "Fast. Cheap. Good. Pick two." "We see a lot of arses on handlebars around here." - [J Ekdahl] |
 |
dkinzer [...] I have just received a PCN notifying a die change which removes this option.
We also came out with a PCN, where we are adding a manufacturing site of the ATmege328 variants. In order to reduce cost, they are removing the full-swing crystal oscillator from these devices too. You will still be able to get the older version of the device, which supports the full-swing crystal oscillator, with a special SL code. At this time, I don't have the SL code.
In my tests with the production mega328PB (date code 1543) I found the now-undocumented full-swing oscillator better behaved than the recommended low-power oscillator at 14.7MHz and 16MHz. Although I haven't researched the problem, I suspect that serial I/O is coupling to the oscillator. The symptoms are that with the low-power oscillator running at 16MHz I couldn't reliably do serial I/O but with the fuses set for the full-swing oscillator (and no other changes) it works reliably.
theusch I feel like I'm watching an old friend slowly waste away...
Indeed. While very few of our designs run faster than 8MHz, we always use the full-swing option in our [mostly industrial] designs. With this change, I wonder if we will see some subtle changes in noisy environments.
I don't have an exact count, but perhaps about 100 production Mega88-family designs over the last 10 years or so. As we run at modest speeds (3.68 and 7.37MHz), our selection of crystal brand and model and load cap value wasn't critical--the designs "just worked". We may well be back to examining ESR and tuning load cap value for best "low-power crystal" operation.
The symptoms are that with the low-power oscillator running at 16MHz I couldn't reliably do serial I/O but with the fuses set for the full-swing oscillator (and no other changes) it works reliably.
The 1284P has had this problem from the get-go. The solution is the same.
With full-swing no longer 'supported' for the new die revisions, this does not bode well.
|
"Experience is what enables you to recognise a mistake the second time you make it." "Good judgement comes from experience. Experience comes from bad judgement." "Wisdom is always wont to arrive late, and to be a little approximate on first possession." "When you hear hoofbeats, think horses, not unicorns." "Fast. Cheap. Good. Pick two." "We see a lot of arses on handlebars around here." - [J Ekdahl] |
|
and funny the datasheet from 11-1015 still talk about full swing osc., (and no warning about some designs gets obsolete )
The real "fun" for Atmel will be for all the hobby people that use mega328 20MHz that can't run 20MHz like the old ones (sure they will understand that they have to buy a 20MHz osc, that cost the same as the chip, and it don't fit the PCB).
and the WC154601 was from 17/7 2015 so someone forgot to tell the schoolkid that made the changes in November, that there was newer info.
Yes I guess I will need to look for other ways, because I use 5V full swing because it's for industrial use. (but I guess that isn't where volume are).,(But I use 324 so perhaps there are a little time left).
or perhaps this in WC154601 tell the truth :
Title: Additional manufacturing location and die revision for selected ATmega48, ATmega88, ATmega168 and ATmega328 microcontrollers
means that it's only for a cheap high volume version, that probably only needs to run 8MHz from internal clk.
But then they should have been given an other name, like the tiny48 is a pricecut version. (yes I know that there are other cuts aswell)
We too use 50K+/yr of the M48-M328's and all have at least a 16M xtal used in industrial settings, and wide temperature swings, so will need to speak to supplier about future availability of the chips with FS xtal osc.
Jim
theusch To OP: Thank you for bringing this to our attention.
glen@proconel.com Atmel, please will you respond to this confusion.
jgmdesign 
I have reached out to a contact I have at Atmel Norway and included a link to this thread.
Let's see what happens.
Jim
theusch Glen, it would be best if you "open a ticket". While a few Atmel people might have an ear here, this is not a support forum so nothing would be official anyway.
The local FAE said this in a recent email:
Quote:We also came out with a PCN, where we are adding a manufacturing site of the ATmege328 variants. In order to reduce cost, they are removing the full-swing crystal oscillator from these devices too. You will still be able to get the older version of the device, which supports the fll-swing crystal oscillator, with a special SL code. At this time, I don't have the SL code.
Wow, usually such a significant change generates a new part code, not done in a hidden manner that will bite the biggest customers the worst.
(wonders just how much a full-swing crystal oscillator actually incrementally costs ? )
Purchasing managers will just enter the same part code, expecting to well, get the same part.
Some candidates for renamed variants :)
Mega328PC 'C' for crippled ? - but sounds like an incremental improvement, to the unwary, so may be best avoided ?
Mega328PN 'N' for nobbled?
Mega328PS 'S' for Stuff removed
Mega328PR 'R' for Reduced Features
Mega328PL 'L' for Lobotomized (reserved for future release)
In my tests with the production mega328PB (date code 1543) I found the now-undocumented full-swing oscillator better behaved than the recommended low-power oscillator at 14.7MHz and 16MHz. Although I haven't researched the problem, I suspect that serial I/O is coupling to the oscillator. The symptoms are that with the low-power oscillator running at 16MHz I couldn't reliably do serial I/O but with the fuses set for the full-swing oscillator (and no other changes) it works reliably.
Cross talk could certainly cause issues, and makes crystal choice trickier.
Usually Lower CL parts would give more gain margin on a rolling-off amplifier, but they will also give more crosstalk disturbance per femtofarad of crosstalk.
Lower ESR crystals should give more amplitude, but the low-power oscillators are usually just current fed N-FETS and there the Vgs more dominates what sets the amplitude.
I have found using a high speed diode (BAW62/ 1N4148) with a simple multimeter can give useful crystal peak indications, with much less disturbance than a scope probe. Works very well for A-B compare testing.
theusch I have found using a high speed diode (BAW62/ 1N4148) with a simple multimeter can give useful crystal peak indications, with much less disturbance than a scope probe. Works very well for A-B compare testing.
I thought about this ('scope probe loading) after writing
We may well be back to examining ESR and tuning load cap value for best "low-power crystal" operation.
As the family in question has a CLKO pin, I thought I'd rig up the test with that enabled. That should at least tell you if the clock is "solid", right, with scope in envelope mode or similar?
But yes, if tuning the load caps ...
As a commercial company we have to change designs, production facility and so on to stay competitive. Clearly any change to an existing device is inconvenient for the users - no discussion about that.
When we send out a PCN it is to inform our customers so that they can act on the change to minimize the impact. In this case we knew that the removal of the full-swing oscillator could be a problem for some customers and for that reason the old revision of the device can be ordered using a special order code. If you need the full-swing oscillator or need to run at more than 16MHz please contact you local sales office (Atmel of distributor): They know the special order codes.
So why don't you write that in the datasheet ? (there have been an update since this was known)
And I have to ask is this the beginning of the end for full swing osc. ?
...or need to run at more than 16MHz...
Wow. So when other manufacturers are releasing new 8-bit commodity devices that run at 72MHz, Atmel are actually taking a step back and slowing down there chips?
jgmdesign Jacob wrote:...or need to run at more than 16MHz...
Wow. So when other manufacturers are releasing new 8-bit commodity devices that run at 72MHz, Atmel are actually taking a step back and slowing down there chips?
I was waiting to see who was going to type that. Good point though.
JIm
dkinzer The oficial Product Change Notice for the 48/88/168/328 is available via the Digi-Key site:
I was waiting to see who was going to type that. Good point though.
Well someone had to say it; we were all thinking it.
As this is an Atmel forum I won't mention the 8-bitter I'm actually having fun with. But they do make the ATmega range look so last century.
ashokok 
If you need the full-swing oscillator or need to run at more than 16MHz please contact you local sales office (Atmel of distributor): They know the special order codes.
will the local sales office or distributor show interest in qty <200 pieces ??
Ashok
If you need the full-swing oscillator or need to run at more than 16MHz please contact you local sales office (Atmel of distributor): They know the special order codes.
Ouch.
The 'special order codes' really should be the new parts, not the old ones.
That's quite poor version control management shown by Atmel.
This also represents a quite insidious failure mode, to many unsuspecting users.
Some may release products thru production testing, that are only just working, to have hard to define field failures.
I take it that means the improvements some have mentioned to the low-swing oscillator, are still not enough, even with ESR and crystal selection care, to reliably reach 20MHz ?
What about Ceramic resonators ? Can 20MHz be spec'd with those, on the low-swing-only parts ?
Other companies have low-swing parts to above 20MHz, so this is clearly not some inherent, unsolvable issue.
Devil's Advocate Says: Would you rather have a Product Change Notice or an End of Life Notice?
Jim
It's not that bad, as long they are sold only as max 16MHz and only for toy use only
(as said before they should have had an extra digit added or so.)
theusch So why don't you write that in the datasheet ? (there have been an update since this was known)
I noticed that there is an Errata item on this for every member of the Mega48 (non-PB) family... rev. J datasheet, 11-2015.
...
...
Does anyone understand what do they mean by "Changed device ID"? "Any die revision", "Previous die revision" - WTF? Whatever used to be A is now PA (starting from rev K)?
ront1234 It's confusing, the PCN posted in #11 does not list a ATmega328p-pu (28 pin dip) so you would think it's not affected by the die change. But they give a general statement that seems to say that all 328's are affected, like in section 40.8 posted above in #35.
clawson 
so you would think it's not affected by the die change. But they give a general statement that seems to say that all 328's are affected
You do realise that if you buy a DIP or a QFP or a BGA or a QFN or an MLF that what's actually inside the plastic (for a particular batch) is exactly the same silicon? They don't have variants for each type of plastic. Similarly they don't make 0..8MHz and 0..16MHz variants (or whatever) separately. They simply grade the production and package them as one or the other. So if they modified the masks for production or switched fabs (and as a consequence the manufacturing geometry) then it is ALL subsequent production that is affected.
I 100% agree with same die in different houses.
But if it's 8 and 16 MHz chips don't necessary are the same chip (A good change it is), many 8 MHz chips are low power, and could be made different, than the faster chips!
(probably same mask but not necessary same thickness of the layers).
And I change the 100% because it looks like some of the newer versions, of some of the chips, won't be in a dip version (make a dip chip an safe way to get an old version!)
ront1234 so you would think it's not affected by the die change. But they give a general statement that seems to say that all 328's are affected
You do realise that if you buy a DIP or a QFP or a BGA or a QFN or an MLF that what's actually inside the plastic (for a particular batch) is exactly the same silicon? They don't have variants for each type of plastic. Similarly they don't make 0..8MHz and 0..16MHz variants (or whatever) separately. They simply grade the production and package them as one or the other. So if they modified the masks for production or switched fabs (and as a consequence the manufacturing geometry) then it is ALL subsequent production that is affected.
They don't list the 328P-PU in the PCN, does this mean another DIP (28P3) package will bite the dust like the newer PB AVRs?
only Atmel know that!
but as long it's not in the data sheet don't expect it to be a normal chip , but if someone ask a millions in DIP I guess they will make them for you, and perhaps even make it a normal chip.
About 15 years ago I was starting a project and started production with AT4414, but it got obsolete but I should get 8515 for the same price, later I was told that Atmel never have made a 4414, it has always been 8515 with an other id, and because the sale was to small they dropped the chip.
Zoinks Scoob... who's steering the ship?!?
In a last minute production decision, Italian Sports car manufacturer Ferrari has decided to remove 3 of the 12 cylinders from its powerhouse F12 Berlinetta.
This surprising move was due to a lathe being down in one of its main facilities, at which time management decided it would be best to outsource production of its engine block.
Ferrari chose to move production to an older facility in Detroit USA, and had to shorten the engine block to fit on the machine that used to make the K-Car.
.... see where I am going with this?
Brad
Ok today I received a email from digikey about the new 328PB (48PB 88PB) etc., so I guess that hell can break out!!!
They totally forget to tell that it can't run 20MHz without an external clk (16MHz is max for a crystal )!!!
see #25 if you buy chips today and run full swing or 20MHz
So what if you want full-swing operation but still no power drain for clock during power-down operation?
I don't see how an external oscillator is going to do this without wild contortions at least.
I'm shocked that Atmel would do something so stupid as to take a critical feature out of an existing design.
I got going with Atmel because of Arduino but what with one thing and another lately I'm going to be looking
at other micros now.
If it's for a new design, I have to ask do you really need to run 20MHz?
For most things 16 or less will get the job done without problems (the internal 8MHz is fine for most things).
