The following parts need special attention:

AT90S1200

The ISP programming protocol of the AT90S1200 differs in subtle ways from that of other AVRs. Thus, not all programmers support this device. Known to work are all direct bitbang programmers, and all programmers talking the STK500v2 protocol.

AT90S2343

The AT90S2323 and ATtiny22 use the same algorithm.

ATmega2560, ATmega2561

Flash addressing above 128 KB is not supported by all programming hardware. Known to work are jtag2, stk500v2, and bit-bang programmers.

ATtiny11

The ATtiny11 can only be programmed in high-voltage serial mode.

Note that some official Microchip programmers store the bitclock setting and will continue to use it until a different value is provided. This applies to "2nd gen" programmers (AVRISPmkII, AVR Dragon, JTAG ICE mkII, STK600) and "3rd gen"programmers (JTAGICE3, Atmel ICE, Power Debugger). "4th gen" programmers (PICkit 4, MPLAB SNAP) will store the last user-specified bitclock until the programmer is disconnected from the computer.

If config-file is written as +filename then this file is read after the system wide and user configuration files. This can be used to add entries to the configuration without patching your system wide configuration file. It can be used several times, the files are read in same order as given on the command line.

reset

The ‘/RESET’ signal will be left activated at program exit, that is it will be held low, in order to keep the MCU in reset state afterwards. Note in particular that the programming algorithm for the AT90S1200 device mandates that the ‘/RESET’ signal is active before powering up the MCU, so in case an external power supply is used for this MCU type, a previous invocation of avrdude with this option specified is one of the possible ways to guarantee this condition. reset is supported by the linuxspi and flip2 programmer options, as well as all parallel port based programmers.

noreset

The ‘/RESET’ line will be deactivated at program exit, thus allowing the MCU target program to run while the programming hardware remains connected. noreset is supported by the linuxspi and flip2 programmer options, as well as all parallel port based programmers.

vcc

This option will leave those parallel port pins active (i. e. high) that can be used to supply ‘Vcc’ power to the MCU.

novcc

This option will pull the ‘Vcc’ pins of the parallel port down at program exit.

d_high

This option will leave the 8 data pins on the parallel port active. (i. e. high)

d_low

This option will leave the 8 data pins on the parallel port inactive. (i. e. low)

Multiple exitspec arguments can be separated with commas.

If avrdude has been configured with libserialport support, a serial port can be specified using a predefined serial adapter type in avrdude.conf or .avrduderc, e.g., ch340 or ft232r. If more than one serial adapter of the same type is connected, they can be distinguished by appending a serial number, e.g., ft232r:12345678. Note that the USB to serial chip has to have a serial number for this to work. Avrdude can check for leading and trailing serial number matches as well. In the above example, ft232r:1234 would also result in a match, and so would ft232r:...5678. If the USB to serial chip is not known to avrdude, it can be specified using the hexadecimal USB vendor ID, hexadecimal product ID and an optional serial number, following the serial number matching rules described above, e.g., usb:0x2341:0x0043 or usb:2341:0043:12345678. To see a list of currently plugged-in serial ports use -P ?s. In order to see a list of all possible serial adapters known to avrdude use -P ?sa.

On Win32 operating systems, the parallel ports are referred to as lpt1 through lpt3, referring to the addresses 0x378, 0x278, and 0x3BC, respectively. If the parallel port can be accessed through a different address, this address can be specified directly, using the common C language notation (i.e., hexadecimal values are prefixed by ‘0x’ ).

For the JTAG ICE mkII and JTAGICE3, if avrdude has been configured with libusb support, port can alternatively be specified as usb[:serialno]. This will cause avrdude to search the programmer on USB. If serialno is also specified, it will be matched against the serial number read from any JTAG ICE mkII found on USB. The match is done after stripping any existing colons from the given serial number, and right-to-left, so only the least significant bytes from the serial number need to be given.

As the AVRISP mkII device can only be talked to over USB, the very same method of specifying the port is required there.

For the USB programmer "AVR-Doper" running in HID mode, the port must be specified as avrdoper. Libhidapi support is required on Unix and Mac OS but not on Windows. For more information about AVR-Doper see https://www.obdev.at/products/vusb/avrdoper.html.

For the USBtinyISP, which is a simplistic device not implementing serial numbers, multiple devices can be distinguished by their location in the USB hierarchy. See the respective Troubleshooting entry in the detailed documentation for examples.

For the XBee programmer the target MCU is to be programmed wirelessly over a ZigBee mesh using the XBeeBoot bootloader. The ZigBee 64-bit address for the target MCU's own XBee device must be supplied as a 16-character hexadecimal value as a port prefix, followed by the ‘@’ character, and the serial device to connect to a second directly contactable XBee device associated with the same mesh (with a default baud rate of 9600). This may look similar to: 0013a20000000001@/dev/tty.serial.

For diagnostic purposes, if the target MCU with an XBeeBoot bootloader is connected directly to the serial port, the 64-bit address field can be omitted. In this mode the default baud rate will be 19200.

For programmers that attach to a serial port using some kind of higher level protocol (as opposed to bit-bang style programmers), port can be specified as net:host:port. In this case, instead of trying to open a local device, a TCP network connection to (TCP) port on host is established. Square brackets may be placed around host to improve readability, for numeric IPv6 addresses (e.g. net:[2001:db8::42]:1337). The remote endpoint is assumed to be a terminal or console server that connects the network stream to a local serial port where the actual programmer has been attached to. The port is assumed to be properly configured, for example using a transparent 8-bit data connection without parity at 115200 Baud for a STK500.

Note: The ability to handle IPv6 hostnames and addresses is limited to Posix systems (by now).

Classic devices may have the following memories in addition to eeprom, flash, signature and lock:

calibration

One or more bytes of RC oscillator calibration data

efuse

Extended fuse byte

fuse

Fuse byte in devices that have only a single fuse byte

hfuse

High fuse byte

lfuse

Low fuse byte

lock

Lock byte

prodsig

Signature, calibration byte and serial number in a small read-only memory, which is only documented to be available for ATmega324PB, ATmega328PB, ATtiny102 and ATtiny104; programmers may or may not be able to read this memory

sigrow

Memory alias for prodsig

usersig

Three extra flash pages for firmware settings; this memory is not erased during a chip erase. Only some classic parts, ATmega(64|128|256|644|1284|2564)RFR2, have a usersig memory. Usersig is different to flash in the sense that it can neither be accessed with ISP serial programming nor written to by bootloaders; avrdude offers JTAG programming of classic-part usersig memories. As with all flash-type memories the -U option can only write 0-bits but not 1-bits. Hence, usersig needs to be erased before a file can be uploaded to this memory region, e.g., using -T "erase usersig" -U usersig:w:parameters.hex:i

io

Volatile register memory; it cannot be accessed by external programming methods only by bootloaders, which has limited use unless the bootloader jumps to the application directly, i.e., without a WDT reset

sram

Volatile RAM memory; like io it cannot be accessed by external programming

ATxmega devices have the following memories in addition to eeprom, flash, signature and lock:

application

Application flash area

apptable

Application table flash area

boot

Boot flash area

fuse0

A.k.a. jtaguid: JTAG user ID for some devices

fuse1

Watchdog configuration

fuse6

Fault detection action configuration TC4/5 for ATxmega E series parts

fuseN

Other fuse bytes of ATxmega devices, where N is 2, 4 or 5, for system configuration

prodsig

The production signature row is a read-only memory section for factory programmed data such as the signature and calibration values for oscillators or analogue modules; it also contains a serial number that consists of the production lot number, wafer number and wafer coordinates for the part

sigrow

Memory alias for prodsig

usersig

Additional flash memory page that can be used for firmware settings; this memory is not erased during a chip erase

io

Volatile register memory; avrdude can read this memory but not write to it using external programming

sram

Volatile RAM memory; cannot be usefully accessed by external programming

Modern 8-bit AVR devices have the following memories in addition to eeprom, flash, signature and lock:

fuse0

A.k.a. wdtcfg: watchdog configuration

fuse1

A.k.a. bodcfg: brownout detection configuration

fuse2

A.k.a. osccfg: oscillator configuration

fuse4

A.k.a. tcd0cfg (not all devices): timer counter type D configuration

fuse5

A.k.a. syscfg0: system configuration 0

fuse6

A.k.a. syscfg1: system configuration 1

fuse7

A.k.a. append or codesize: either the end of the application code section or the code size in blocks of 256/512 bytes

fuse8

A.k.a. bootend or bootsize: end of the boot section or the boot size in blocks of 256/512 bytes

fusea

A.k.a. pdicfg: configures/locks updi access; it is the only fuse that consists of two bytes

fuses

A "logical" memory of up to 16 bytes containing all fuseX of a part, which can be used to program all fuses at the same time

osc16err

Two bytes typically describing the 16 MHz oscillator frequency error at 3 V and 5 V, respectively

osc20err

Two bytes typically describing the 20 MHz oscillator frequency error at 3 V and 5 V, respectively

osccal16

Two oscillator calibration bytes for 16 MHz

osccal20

Two oscillator calibration bytes for 20 MHz

prodsig

Read-only memory section for factory programmed data such as the signature, calibration values and serial number

sigrow

Memory alias for prodsig

sernum

Serial number with a unique ID for the part (10 or 16 bytes)

tempsense

Temperature sensor calibration values

bootrow

Extra page of memory that is only accessible by the MCU in bootloader code; UDPI can read and write this memory only when the device is unlocked; bootrow is not erased during chip erase

userrow

Extra page of EEPROM memory that can be used for firmware settings; this memory is not erased during a chip erase

sib

Special system information block memory with information about AVR family, chip revision etc.

io

Volatile register memory; avrdude can program this memory but this is of limited utility because anything written to the io memory will be undefined or lost after reset; writing to individual registers in the terminal can still be used, e.g., to test I/O ports

sram

Volatile RAM memory; can be read and written but contents will be lost after reset

The op field specifies what operation to perform:

r

read device memory and write to the specified file

w

read data from the specified file and write to the device memory

v

read data from both the device and the specified file and perform a verify

The filename field indicates the name of the file to read or write. The format field is optional and contains the format of the file to read or write. Format can be one of:

i

Intel Hex

I

Intel Hex with comments on download and tolerance of checksum errors on upload

s

Motorola S-record

r

raw binary; little-endian byte order, in the case of the flash ROM data

e

ELF (Executable and Linkable Format)

m

immediate mode; actual byte values are specified on the command line, separated by commas or spaces in place of the filename field of the -U option. This is useful for programming fuse bytes without having to create a single-byte file or enter terminal mode.

a

auto detect; valid for input only, and only if the input is not provided at stdin.

d

decimal; this and the following formats generate one line of output for the respective memory section, forming a comma-separated list of the values. This can be particularly useful for subsequent processing, like for fuse bit settings.

h

hexadecimal; each value will get the string 0x prepended.

o

octal; each value will get a 0 prepended unless it is less than 8 in which case it gets no prefix.

b

binary; each value will get the string 0b prepended.

When used as input, the m, d, h, o and b formats will use the same code for reading lists of numbers separated by white space and/or commas. The read routine handles decimal, hexadecimal, octal or binary numbers on a number-by-number basis, and the list of numbers can therefore be of mixed type. In fact the syntax, is the same as for data used by the terminal write command, i.e., the file's input data can also be 2-byte short integers, 4-byte long integers or 8-byte long long integers, 4-byte floating point numbers, 8-byte double precision numbers, C-type strings with a terminating nul or C-like characters such as ''. Numbers are written as little endian to memory. When using 0x hexadecimal or 0b binary input leading zeros are used to determine the size of the integer, e.g., 0x002a will occupy two bytes and write a 0x2a to memory followed by 0x00, and 0x01234 will occupy 4 bytes. See the description of the terminal write command for more details.

In absence of an explicit file format, the default is to use auto detection for input files, and raw binary format for output files. Note that if a filename contains a colon as penultimate character the format field is no longer optional since the last character would otherwise be misinterpreted as format.

When reading any kind of flash memory area (including the various sub-areas in Xmega devices), the resulting output file will be truncated to not contain trailing 0xFF bytes which indicate unprogrammed (erased) memory. Thus, if the entire memory is unprogrammed, this will result in an output file that has no contents at all. This behaviour can be overridden with the -A option.

As an abbreviation, the form -U filename is equivalent to specifying -U flash:w:filename:a. This will only work if filename does not have a pair of colons in it that sandwich a single character as otherwise the first part might be interpreted as memory, and the single character as memory operation.

data can be binary, octal, decimal or hexadecimal integers, floating point numbers or C-style strings and characters. If nothing matches, data will be interpreted as the name of a file containing data, which will be read and inserted at this point. In order to force the interpretation of a data item as file, e.g., when the file name would be understood as a number otherwise, the file name can be given a :f format specifier. In absence of a format suffix, the terminal will try to auto-detect the file format.

For integers, an optional case-insensitive suffix specifies the data size: HH 8 bit, H/S 16 bit, L 32 bit, LL 64 bit. Suffix D indicates a 64-bit double, F a 32-bit float, whilst a floating point number without suffix defaults to 32-bit float. Hexadecimal floating point notation is supported. An ambiguous trailing suffix, e.g., 0x1.8D, is read as no-suffix float where D is part of the mantissa; use a zero exponent 0x1.8p0D to clarify.

An optional U suffix makes integers unsigned. Ordinary 0x hexadecimal and 0b binary integers are always treated as unsigned. +0x, -0x, +0b and -0b numbers with an explicit sign are treated as signed unless they have a U suffix. Unsigned integers cannot be larger than 2^64-1. If n is an unsigned integer then -n is also a valid unsigned integer as in C. Signed integers must fall into the [-2^63, 2^63-1] range or a correspondingly smaller range when a suffix specifies a smaller type.

Ordinary 0x hexadecimal and 0b binary integers with n digits (counting leading zeros) use the smallest size of one, two, four and eight bytes that can accommodate any n-digit hexadecimal/binary integer. If an integer suffix specifies a size explicitly the corresponding number of least significant bytes are written, and a warning shown if the number does not fit into the desired representation. Otherwise, unsigned integers occupy the smallest of one, two, four or eight bytes needed. Signed numbers are allowed to fit into the smallest signed or smallest unsigned representation: For example, 255 is stored as one byte as 255U would fit in one byte, though as a signed number it would not fit into a one-byte interval [-128, 127]. The number -1 is stored in one byte whilst -1U needs eight bytes as it is the same as 0xFFFFffffFFFFffffU.

One trailing comma at the end of data items is ignored to facilitate copy & paste of lists.

It is quite possible, as is with direct writing to the underlying fuses and lock bits, to brick a part, i.e., make it unresponsive to further programming with the chosen programmer: here be dragons.

Option -a displays the register I/O addresses in addition; -m displays the register memory addresses used for lds/sts opcodes instead of the I/O addresses. Option -s also shows the size of the register in bytes whilst -v shows a slightly expanded register explanation alongside each register.