# crystals for atmel chips

a newb question----

i have some atmega644pa-au chips- datasheet states:  Speed Grades ̶ 0 - 4MHz @ 1.8 - 5.5V ̶ 0 - 10MHz @ 2.7 - 5.5V ̶ 0 - 20MHz @ 4.5 - 5.5V

does that mean that there are only the 3 speeds for external or is it possible to use a 16MHz???

aes92000: a newb question----

i have some atmega644pa-au chips- datasheet states:  Speed Grades ̶ 0 - 4MHz @ 1.8 - 5.5V ̶ 0 - 10MHz @ 2.7 - 5.5V ̶ 0 - 20MHz @ 4.5 - 5.5V

does that mean that there are only the 3 speeds for external or is it possible to use a 16MHz???

This is telling you that the processor can be clocked at upto 4mhz with a 1.8V power supply. Up to 10mhz needs at least 2.7v and 20mhz needs Vcc to be at least 4.5v. The 5.5v is the maximum the processor can safely handle.

So to use 16mhz crystal you need Vcc to be higher than 2.7v I use 3.3v with 16mhz resonator on a 328p.

Chuck.

chucktodd: I use 3.3v with 16mhz resonator on a 328p.

Have you ever had a problem over clocking it like that?

Those are marked 0 - 20, not 20mhz - as in, it's a range, from 0mhz to 20mhz, and anything in that range is within spec.

chucktodd: So to use 16mhz crystal you need Vcc to be higher than 2.7v I use 3.3v with 16mhz resonator on a 328p.

Chuck.

See page 310 of the datasheet:

"As shown in Figure 29-1, the Maximum Frequency vs. VCC curve is lin- ear between 1.8V < VCC < 2.7V and between 2.7V < VCC < 4.5V."

10mhz @ 2.7, 20mhz @ 4.5

4.5-2.7=1.8v 20-10=10mhz

1mhz per 0.18v

That implies that 16mhz is only within spec at 2.7+6*(0.18)= 2.7+1.08 = 3.78v or higher.

That said, it's widely reported that the chips work at 16mhz @ 3.3v, despite this being out of spec. Might not work over the whole temperature range or something, though.

jboyton: Have you ever had a problem over clocking it like that?

nope.

Here is the DigiKey partnumber 490-1198-1-ND 16mhz resonator. I connect a 1M resistor in parallel. I have not had any problems.

chuck.

Is there a reason you use the resonator instead of a crystal?

DrAzzy: Is there a reason you use the resonator instead of a crystal?

Acceptable accuracy, cheap price(\$0.32 per 100), fewer parts(no Caps needed)

Chuck.

I use these http://www.dipmicro.com/store/XC7-16000 less than a quarter each, and two of these, less than 7 cents each, http://www.dipmicro.com/store/C1K22-50 No problems. Stock up, you'll use them in all your projects.

CrossRoads: I use these http://www.dipmicro.com/store/XC7-16000 less than a quarter each, and two of these, less than 7 cents each, http://www.dipmicro.com/store/C1K22-50 No problems. Stock up, you'll use them in all your projects.

Wow, those HC49's are big. The resonators I use are 3.5 x 1.5 x0.6mm. Those HC49's are 11.5 x 4.7 x 3.5mm, bigger than the 328 TQFP!

Chuck.

Hard to manufacture a small 16Mhz xtal that are cheap. Small solid state devices are easy. I prefer the stability of a crystal over a resonator.

.

chucktodd:
nope.

Here is the DigiKey partnumber 490-1198-1-ND 16mhz resonator. I connect a 1M resistor in parallel. I have not had any problems.

chuck.

why the resistor in parallel?

LarryD: Hard to manufacture a small 16Mhz xtal that are cheap... I prefer the stability of a crystal over a resonator.

imo silly to consider ceramic over quartz in this day and age. unlike decades ago quartz is much cheaper and far more accurate. on world market hc49s 2-3 cents ea and 16mhz watch crystal size or 5x3.2mm smd less than a dime. sometimes as low as a nickel.

personally i never uses any caps unless absolute accuracy required. in that case tcxo best bet anyway.

DaveEvans: why the resistor in parallel?

page 3:

The feedback resistor increases the probability that the resonator starts oscillating.

Chuck.

chucktodd: The feedback resistor increases the probability that the resonator starts oscillating.

The Uno R3 has this resistor. But what is the probability that it will fail to start oscillating without the 1M resistor?

By the same token overclocking a little increases the probability of the processor failing. But apparently not by much.

jboyton: The Uno R3 has this resistor. But what is the probability that it will fail to start oscillating without the 1M resistor?

By the same token overclocking a little increases the probability of the processor failing. But apparently not by much.

My projects have never had a clock failure, so I cannot give you a probability.

Chuck.

Without knowing how much including that resistor increases the probability that the resonator will oscillate it’s hard to calculate if it’s even worth bothering. I think it’s similar to straying above the guaranteed speed limit of the processor.

jboyton: Without knowing how much including that resistor increases the probability that the resonator will oscillate it's hard to calculate if it's even worth bothering. I think it's similar to straying above the guaranteed speed limit of the processor.

The reason I do it, is because the resonator manufacturer recommends it. A single 0402 1M resistor next to the resonator is an acceptable use of board space.

Chuck.

I understand. It's easier to add a resistor that doesn't matter 99.99% of the time than it is to increase Vcc by at least 0.5V or reduce the clock frequency by at least 3MHz -- even though that doesn't matter 99.99% of the time either.

maybe worth noting that avr chips, like most mcu families, have an internal resistor. i have measured from 600k to 1.2m depending on particular device. considering input impedance of the cmos inverter one 100x higher would still be adequate. in fact even if there were not an internal one then resistance from the epoxy package guarantees proper operation.

also note that some like m8 have internal caps too which can be enabled. personally i dont bother with internal or external caps as there is little effect on startup, stability, or accuracy.

In the Atmel app note AVR042 this representation of the oscillator circuits is provided:

In the text it says that Rf is "approximately 1MΩ" but it seems they are talking about the second circuit (B), the one for the low frequency oscillator. Both drawings include an Rf so it may be applicable to the higher frequency crystal / resonator oscillator as well.

Can you just attach a DMM to the oscillator pins while the chip is sitting on the table and get a meaningful value?