Ref_PartSelection.md

AVR device selection guide

There are three modern tinyAVR series, and within each family, a dozen or more parts across 3-4 different pincounts.

This document, I hope, will help you determine the best part for your needs.

Most prominent differences

This table lists some of the most significant differences between the tinyAVR 0, 1, and 2 series parts and the mainline AVR Dx-series and tinyAVR 2-s

1+series denotes the 5 16k or 32k parts in the 1-series. These have significantly enhanced features, without an accompanying jump in price. Two recent non-tinyAVR are shown for compariso. The DD-series is price-competitive with the non-tinies. The EA-series looks like its going to be a bit pricier. The AVR DA/DB are not included in this chart - these parts are aiming for lower cost and the right size and selection of peripherals for typical applications. The DA/DB series was "Okay, we scuttled our old flagship, what's next?" "Now we get aboard our new one" .... "Well, you heard the man! Climb aboard and start bailing!" "Maybe we should have made sure it was seaworthy first?" "As if we had time for that! "

Parameter	0-series	1-series	1+series	2-series	DD-series	EA-series	EB-series
Flash sizes	2-16k	2-8k	16-32k	4-32k	16-64k	16-64k	16-32k
RAM	128b-1k	128b-1k	2k	256b-3k	2k-8k	2k-6k	2k-3k
Pincounts	8-24	8-24	14-24	14-24	14-32	28-48	14-32
CCL LUTs	2	2	2	4	4	4	4
EVSYS chan	1syn/2asyn	2syc/4asy	2sy/4asy	6	6	6 v1.2	6 v1.2
ADC type	Normal	Normal	2xNormal	Differential w/ 1/2/4/8/16x gain	Semidifferential	Differential w 1/2/4/8/16x gain, ADC specs better.	<-- same as EA
ADC res	10-bit	10-bit	10-bit	12-bit	12-bit	12-bit	12-bit
ADC speed	4-6x classic	4-6x classic	4-6x classic	14x classic	5-8x classic	14x classic	14x classic
VREF opts	Dumb	Dumb	Dumb+ext	Good	Good	Good	Good
DACs	0	1x8b v0.9	1x8b v0.9	0	1x10b v1.0	1x10b v1.1	None
AC's	1	1	3	1	2	2	2
AC DACREF	No	Yes*	Yes*	Yes	Yes	Yes	Yes
USARTs	1	1	1	2	2	2-3	1
Alt Reset	No	No	No	20/24-pin	Irrel.	Irrel.	Irrel.
TCAs	1	1	1	1	1	2	No
TCBs	1x v1.0	1x v1.0	2x v1.0	2x v1.1	2/3	2-4	2
TCD0	No	Yes	Yes	No	Yes	No	No
TCE0	No	No	No	No	No	No	Yes + WEX
TCFs	No	No	No	No	No	No	1
Max pract. internal overclock	25 MHz	25 MHz	32 MHz	40-48 w/xtal	Basically none.	TBD - potentially a lot
Max pract. overclock	30 MHz	30 MHz	30 MHz	32 MHz	40-48 w/xtal	Expect 32-40	TBD, but no xtal
Errata	A lot	Tons	Tons	Not much	Hardly any?	Virtually none, Except for nvmctrl, that one's amess	We'll see...
Price/pc	$1.08	n/a	$1.16	$1.18	$1.10	expect $1.20	Expect, uh between $1.00 and $1.30?

Qty 1 price from Microchip Direct for 16k flash version of 24 pin tiny, normal temp range, or 28-pin DD or EA in QFN (since it is not being made in low pincounts). Put simply, the price difference is minuscule (except maybe for the EA).

2-series TCBs are the only ones that can do cascade mode (not very important) and count on event (if you need it, you really need it). The Dx and Ex-series and EB also got the new version. The TCA is also improved on the 2-series with an extra event input.

* - AC0 DACREF on 1-series is DAC0, the same one that can be output to a pin. This is occasionally inconvenient.

The ADC Reference Voltages

There are two sets of VREF options, one of which is good, and one of which is dumb. Reference voltage | Dumb system | Smart System | notes | Lowest | 0.55V | 1.024V | ADC clock may need to be made much slower. | Low | 1.1V | 2.048V | Note that most classic AVRs have VREF = 1.1V or sometimes an option of 2.56V | Mid | 1.5V | - | Yeah, the better set of references dropped and option and is still better | High | 2.5V | 2.500V | Only option the same on both | Highest | 4.3V | 4.096V | | VDD | VDD | VDD | | External VREF | <VDD | <VDD | See elect. characteristics. |

External vref supported on the 1+series and 2-series. I don't know what numbnuts picked those voltages for the 1-series references. They could have made them less-round numbers or something I guess, even less convenient... but these are intensely inconvenient voltages to have to choose from. Particularly compared to the good ones on the 2-series and every other modern AVR:

1V024. 2V048. 2V5. 4V096, VDD, external.

What's special about these? Mainly the fact that we have an ADC with a native resolution of 12 bits. That's 0-4095 or on Dx-series, 10 bits (1024). So 4V096 reference used for a 12-bit measurement with no gain will return a numeric voltage that is simply equal to the voltage, in millivolts. With a 10-bit ADC, either multiply by 4 for millivolts or do a 12-bit oversampled reading) Hence you can read a pin, say you get back 2360. You want to make that into a human readable voltage? Easy, just print it, followed " mV" to the console or display, and save the grindingly slow division. Other combinations of those three power-of-2 divided by 1000 references can also be super-convenient in various circumstances, because you always end up with the numeric units of your analog reading as being a power of two (positive or negative power, if you say, oversample and decimate to get 16-bits using the 4V096. you end up with a number from 0-65535, in units of 1/16th of a millivolt. That means you can separate the fractional portion with this kind of thing:

/* global scope */
const uint8_t sixteenths[]={0, 6, 13, 19, 25, 31, 38, 44, 50, 56, 63, 69, 75, 81, 88, 94};
/* Now when you take the reading */

uint8_t fract = reading & 0x0F; // 0-15
reading >>= 4; // now this has the whole millivolt part!
Serial.print(reading);
Serial.print('.');
uint8_t temp = sixteenths[fract];
if (temp < 10) {
  Serial.print('0');
}
Seriial.println(temp);

With a 16-byte lookup table, you've managed to drop several expensive divisions and avoid having to use the grindingly slow, flashhogging floats (if you use no floats in your sketch, you save a considerable amount of flash) Note that it was an intentional decision to only show two significant figures - you only have data enough to show with granularity of 16ths). The new references are much much more convenient!

The errata on the 2-series is not bad at all - a couple of the widespread ones we're all used to, plus you need to turn the ADC off manually before going to sleep. The AVR DD-series is also looking pretty good (though I I've come across 3 closely related areas where behavior doesn't match up with expectactions)

The errata on the 0/1-series is a totally different matter. The table of contents is too big to fit on one screen. The RTC and PIT both misbehave horribly if you enable only one of them, the CCL's D-latch doesn't work, and there's a huge variety of other bugs (most of which are worked around by the core).

Plus, the PORTMUX and EVSYS look like a beta-test version. EVSYS channels are divided into two types, the headers didn't give you names for the event users (though I think I papered over this, because it was just so fucking ugly), and the layout of generators across the channels is weird and inconsistent. The Event library makes this a little easier to deal with, but there's no way around the fact that the event system is fiddlier and less powerful. PORTMUX works, it's just unpleasant, because they hadn't thought of naming the registers after what peripherals they controlled the route for yet, so they're CTRLA, CTRLB, CTRLC, CTRLD, instead of TCAROUTEA, USARTROUTEA etc.

The tiny1+ (16k and 32k parts) have a second fully user accessible ADC

On the other hand, the 1+series parts are the only AVR family to feature TWO ADCs and expose both of them as a normal ADC (the Dx has A has a second, but it's only used for PTC). I suspect this was intended to be an escape hatch to be able to simulate a differential ADC (and they already had the second ADC there for the PTC) - but it is much more than that. Almost makes me sad to see that they're using a better ADC on new parts now. Almost. You can:

• Have the two ADCs each monitoring a different channel. I was able to get these parts to continually read two channels with an ADC, alternating between them, and between ADC reads, use a lookup table to convert the ADC reading to a value to send over SPI to a DAC... Both ADCs were free-running, and I'd poll them until the result was ready, lookup, send, then poll other adc, and when i was done optimizing, it was running at the speed of the ADCs - I was not missing a single reading from either ADC. That was a fun project. Only time I think I've ever had fun with an analog input. The overall goal was a 2-channel device to take 2 input voltages, and for each ADC reading, look up an arbitrary voltage to set the DAC to, in order to transform the voltage curve from analog input sources which had one characteristic voltage vs sensor value into the voltage curve that the other device was expecting. Client had no control over the programming on either side (and the "correct" part was already unavailable , and needed a thing to go in the middle to act as a translator.
• Use one ADC normally. Leave the other one sitting in free-running mode with the Window Comparator configured to generate an interrupt in the event of a voltage excursion on a specific pin - without having to take your other ADC's attention away from what it's measuring (remember that the sampling capacitor holds a small but non-zero amount of charge, which is why one usually throws away the first reading when changing channels)
• But you probably don't want to do the simulated differential ADC thing, because you could get a 2-series with a real ADC.

And it has three analog comparators (what?! three?!), complete with the windowed mode like you can do on the AVR DA and DB parts (so these have two completely different ways of window analog comparating both with windowed mode, and with the dual ADC, if you for some reason wanted to, you could compare three voltages against a window entirely in the background, with no CPU involvement other than responding to the interrupt when your specified conditions occurred)

And the memory layout

Parameter	0-series	1-series	1+series	2-series	DD-series	EA-series	DA/DB
128k flash	No	No	No	No	No	No	Yes
ram @ 128k	-	-	-	-	-	-	16k
64k flash	No	No	No	No	Yes	Yes	Yes
ram @ 64k	-	-	-	-	8k	6k	8k
32k flash	No	------->	20/24 pin	Yes	Yes	Yes	Yes
ram @ 32k	-	------->	2k	3k	4k		4k
16k flash	Yes	------->	Yes	Yes	Yes	Yes	No
ram @ 16k	1k	------->	2k	2k	2k	2k	-
8k flash	Yes	Yes	<------	Yes	No	Yes	No
ram @ 8k	512b	512b	<------	1k	-	-	-
4k flash	Yes	Yes	<------	Yes	No	No	No
ram @ 4k	256b	512b	<------	512b	-	-	-
2k flash	8/14 pin	8/14 pin	<------	No	No	No	No
ram @ 2k	128b	512b	<------	-	-	-	-

Okay... So now, let's just go through when you would want to use which series

2-series

You definitely want to use the 2-series if:

• You need a really good ADC
• You need a second serial port
• You you want to use them as the vessel for complex CCL-based schemes. The 2-series parts are the best parts if all you really care about is the CCL-to-price ratio. You can get 4 LUTs for cheaper than on any other part, and on top of that, the pinmapping (while kinda weird) will get you inputs on every channel except LUT0 IN0 (that's the UPDI pin) and an output from every LUT. I've used 424s on one occasion because I determined that what I needed was a TCA, a TCB, 4 CCL luts with a way to get an output pin, and the binary was compiling to around 2k (though it would grow to a hair over 3 by the end). Tiny 424 checks off every box, and SOIC-14 is the kind of package you could solder with ease even after drinking a six-pack of beer (not that I recommend soldering drunk, I just mean, they are one of the easiest surface mount packages to solder, and I consider them faster and less tedious to solder than a DIP).
• If you want autoreset and don't want to have to fart around with HV programming.
• If you like to live on the edge and for some reason need to clock a tinyAVR at 32 MHz and can't imagine why that need for speed might point to some questionable decisions elsewhere in your planning.
• If you don't need anything specific to the 1-series, the 2-series devices are all around better parts. The tiny 0/1-series kinda shipped before they ought to have been (probably because they felt they had to get some new parts out because the classic AVRs were starting to look pretty crude by the standards of the day. That's the decision I would have made too, so I can't knock them for it. Their EVSYS is particularly awful. Event.h tries it's best to paper over it, but it's kinda hard to hide.

1-series

Sometimes, though, the 1-series can do something, and the 2-series can't.

• If you need more than 6 PWM channels. The 1-series has a type D timer that gives you 2 more, and the TCD can do some things the other timers can't.
• If you need the functionality of a TCD.
• If two independent ADCs are more useful to you than one really fancy one.
  • Maybe you need to alternately measure two signals, but their impedance is too high, and each one throws off the reading of the other.
• If you need multiple AC's or analog comparator windowed mode. Note that you can also use the ADC in freerunning mode with the window comparator enabled, and use that instead, and if you want to compare two voltages, on a 2 series you can use differential mode too! But if you need to be using the ADC at the same time, that's not an option.
• If you need to have two window comparators of one form or another (either via AC or via ADC) running at simultaneously
  • In fact, you can have a pair of AC's set up as one window comparator, and then both ADC's operating in free running mode ALSO doing window comparison, if you for some strange reason need to have three windows continuously compared (this is something that I do not know of any other AVR that can do )
• If you need a DAC output on a tinyAVR (but consider a DD too)
• If you need the wacky externally controlled RS485 mode that they removed from the Dx and later (note that it may well still be present, just not documented),

0-series

Uhhhhh.....

• If you can't get anything else.
• If you are more of a tightwad than Ebeneezer Scrooge and the suggestion of spending 2-10 cents more for a 1-series makes you launch into a tirade about how people today don't understand the value of a dollar. Or in this case, a nickel. (Some of the 0-series are actually priced higher than their strictly better 1-series or 1+series counterparts.
• If you are a masochist, and enjoy the pain of working with what appear to be intentionally kneecapped parts.
• Yeah, that's about it. The 0-series tinies are strictly worse than the 1-series on every conceivable metric... and they are practically the same damned price.

The problem with the 0-series is that they have features like a bare-bones budget part, but that's how the 1-series was priced already, so there wasn't much room left to cut prices further. My guess is that the 0-series also doesn't cost much less to manufacture. They may have been expecting greater cost savings than they were able to get. I suspect that the majority of the silicon die area (and hence cost) on an AVR is the AVR core itself, and the memory arrays, and not so much the peripherals.

Don't forget - there are also AVR Dx and Ex-series invading some of the same spaces as tinyAVR

AVR DD-series parts come in 14, 20, 28 and 32-pin packages, with up to 64k flash/8k ram (or as little as 16k and 2k), packaged in SOIC-14, SOICW-20/28, VQFN-20/28/32 (same 3mm x 3mm that tinies got for VQFN-20 - but not available for 64k parts). SSOP-28, DIP28 and TQFP-32. Their errata list only has three issues that hardly anyone will care about. They run faster, use less power, aren't that much more expensive... oh and they've got MVIO too.... Yeah they pretty much blow the doors off the tinyAVRs a most contests - though the tinies still hold onto some of their advantages (CCL pins) , and the EBs only have a single UART. The only real thing the tinies can hold up as an advantage is that the 14 and 20-pin tinies do better at maximizing the availability of peripheral functions on pins, while the DD-series instead maintains a coherent and consistent layout, at the cost of some flexibility at the low pincounts - the one thing conspicuously absent is that fancy new ADC like the 2-series has, which won't reach 14-pin parts from the "double-letter" AVRs until the EB-series, though it comes to 28, 32, and 48-pin packages sooner in the EA-series. EA-series armies will be reinforcing their positions within the tinyAVR highlands, with flash from 8k & 1k ram to 64k and 6k, as the EB-series masses for a full scale invasion with 14, 20, 28 and 32-pin parts, armed with two new waveform generating timer types, and sized at the decidedly tinyAVR-like 8k flash/1k ram to 32k flash 3k ram; they look like they're aiming for a low price point with those for sure, with those kind of memory sizes and pincounts, and only 1 UART.

DD-series

Use a DD series if:

• You'd otherwise use a tiny, except you need the expanded peripheral options, port INLVL setting, MVIO, or the ability to use an HF crystal.
• You want maximum portability of code to other Dx-series parts.
• 20 MHz isn't fast enough, but 24 MHz would be, and you don't like overclocking.
• 30 MHz isn't fast enough, but 40 MHz probably would be and 48 would be for sure - and you're cool with overclocking.

The others

DA/DB series

Use the DA or DB series if:

• Do you really need to ask this? The reasons are pretty obvious: They've got more pins, ram, flash and peripherals than other options (and they're more expensive), but they have more errata than DD or tiny2, or EA.
• You need more pins, flash, ram, or instances of peripherals.
• ADC is 12-bit,
• If you need or want onchip opamps. Only the DB's have OPAMPs.

The EA-series

The EA series boasts a new and improved ADC, some nice improvements that are hidden from the user but will make Event work better. Oh, and RWW flash. They should have made the event system changes long ago, this is the bloody obvious solution... and the clock system has regressed back to tinyAVR, EXCEPT IT KEPT THE AWFUL CAL REGISTER OF DX! Reading that register description felt like a kick to the groin the first time I saw it. I figured that "okay, if we're not getting 24 MHz rated with insane overclockability, we'll at least get the tinyAVR calbyte" - the tinyAVR 2 series could output timing data that would not vary from a quadratic best fit line enough to be seen, and the magnitude of the difference between min and max speed... was about the nominal speed. 0/1-series could be adjusted to 32 MHz but weren't stable unless extenrnal clock and particularly good decoupling was used running from solid 5v supply. The 2-series was generally stable at internal 32 MHz and the tuning ran all the way up to 36ish on the high oscillator.

But nope. No fun for us here. Only way to overclock is with a crystal. And you can use a watch crystal to correct the high speed clock.

But anyway - Use them if you need the new ADC, which is what the 2-series got with fewer rough edges and much larger selection of connections for it to stretch it's muxes to, basically. Don't use them if you don't need that. Dx's are better parts in general.

The EB-series

A mysterious part, this doubles down on swing towards the lower end of the market, after the high end got the DA/DB. It's not as large as a DD (fewer features, slower processor). It has two interesting things - it's the low pincount Ex with the good ADC. That can't be bad! It replaces the TCA with a more complicated and powerful 4-channel 16-bit one - but this one doesn't have split mode :-(. TCF normally is for pulse and frequency ("frequency" = 50% duty cycle) with great precision using it's PLL-clocked (up to 80 MHz rated) TCF for very accurate periodic pulses or frequencies. But who wants to do THAT?! We want PWM! The TCF has a TCB-style "lobotomized mode" in which it generates 2 PWM channels

The DU-series

Hasn't been seen much in marketing material, but it is believed that it is still in development and that the old brief gave generally correct information, hence we have 20/14 pin packages up to 32k, with 64k sizes only on the 32 and 28 pin parts, and yes on the 28-pin qfn (product designers and accountants cheer, everyone else glances around confused), Yes on 14-pin SSOP, and yes on 20-pin 3mmx3mm. Use these parts if you need native usb for something. USB is a demanding master and he requires great sacrifices to be made to secure his blessing. The TCD, MVIO and the exrta TCB some parts went were all thrown before the almighty USB. The entire point of these parts is the native USB. They should not be used if the native USB is not being used.

So I'm going to say you should use the DU-series after it's been released, IFF you are using native USB as in symbolic logic, if and only if)

One notices a pattern with their recent releases

Count how many parts were released in each family since 2020. The AVR DA and DB got 11 parts each (4 pincounts 3 memory sizes, minus the 32k/64-pin). The tinyAVR 2-series got 12 parts (3 pincounts, 4 memory sizes). DD got 12 parts (4 pincounts, 3 memory sizes) The EA-series gets 12 parts (3 pincounts 4 memory sizes) and the EB will be 12 (4 pincounts 3 memory sizes). Even the DU, whose product brief appeared only for a few hours (it was a very brief) listed 11 parts (4 pincounts and 3 memory sizes, and no 14 pin/64k version). It would appear that Microchip Doctrine is to release parts in squads of a dozen, or whatever the opposite of a baker's dozen is (bakers dozen is 13, 1 more than a dozen. I guess a chip baker's dozen is 11.).

My belief is that there are 6 DA/DB dies (1 per flash size (the different flash sizes initial release die rev was different for all three sizes), and the two big pincounts and 2 small pincounts share a die. The 2-series I think is 3 dies - 32k, 16k and 8/4k, while the DD is 4 (16/32k and 64k, and again the two big and two small parts share a die). EA I predict will use 1 die across all pincounts, and end up with either 3 or 4 dies depending on whether 16 and 8 are made with the same die or not. EB and DU will likely both be like the DD in this regard.

These patterns break down when you look further back to the first modern AVRs (0/1-series tinies and mega0) - but those were the chaotic days right after the 2016 revolution. They came extremely close to releasing a 3214 - even had headers for it in the ATpack, then removed them.