DC Supply rails in Embedded projects

This is an important decision and poor choices may lead to poor performance or even device failures. Also an over-complicated supply configuration will add to system complexity. For example consider that the PC-AT power supply "standard" calls for +12V, +5V, -5V and -12V (though the negative rails are low current and the newer ATX standard has quietly depreciated them). This makes for an elaborate power harness. In contrast a typical development board runs off a single supply voltage, often either 5V via a USB connector or 12V via a "barrel" DC connector. Other required supplies are derived by circuits on the board.

Summary of supply rail naming

By convention supply rails are often named with a large "V" followed by a repeated letter, often lower-case or in subscript. Examples: Vcc, Vdd, Vss. They are also sometimes written all-capital: VCC, VDD, VSS.

There's a reason for the double letter convention, I believe it comes from an IEEE paper but the convention has stuck. Some of the labels have changed their meanings and using two names for the same supply can cause conflicts in Schematic CAD tools.

An alternative convention for specific voltages is to write the voltage in decimal then replace the decimal point with a "V", thus 3V3 or 5V0. This matches a convention used for component values. Replacing the decimal point with the unit reduces the risk of the value being misread.

Vcc and GND

These labels are usually reserved for a 5V supply. The C stands for collector as this was originally used for TTL designs. You might get away with labelling a 3.3V supply Vcc but this could cause misunderstandings so be sure to document it to avoid confusion. It is common practice to see the supply pins of HCMOS devices labelled as Vcc and GND, and in schematic CAD packages that hide the supply rails of logic gates their supply pins will automatically be connected to Vcc and GND.

Vdd and Vss 

Originally the Drain and Source supplies of a NMOS design. Older devices had a 12V VDD, an additional VCC pin for a 5V supply, a VBB (below) and VSS meaning ground. Newer devices had a 5V VDD meaning that VDD and VSS were often synonyms for VCC and Ground, and since many CAD packages object to connecting supply symbols together this has been causing problems with design rule checking for at least two decades.

The terms VDD and VSS persist in 4000-series CMOS devices even though in CMOS both supplies go to the source of a FET.

VBB

A negative supply, the B probably stood for "bias" not "base". Apparently this was used for a reference voltage for an ECL (emitter coupled logic) device. It was also used for the -5V supply to a 4116 dynamic RAM chip and to other early NMOS devices. This term has fallen into disuse.

VEE

Another misleading name, it is found on a small number of CMOS devices mostly switches and some LCD drivers. VEE is an optional negative supply (ground if not used) that allows the IC to switch signals below ground. The E probably once stood for emitter but it is now just a label. For an example consider the CD4541 analogue switch. It has regular CMOS supply pins so if powered from 5V and ground it can switch signals in the 0-5V range, but if supplied an additional -5V VEE supply it can switch both positive and negative signals.

VEE is also sometimes used for a negative OUTPUT on a LCD module

The name VEE probably originated with ECL (emitter coupled logic) which in its original form had a positive ground and negative 5.2V supply. 

VO (single "O") is often used for the contrast voltage to a LCD module. This may need to be adjustable.

Keep it simple with a single 3.3V supply

Many current generation microcontrollers and embedded peripherals have a maximum supply voltage of 3.3V, so an all 3.3V design makes sense. Many USB-powered development boards regulate the nominal 5V bus supply down to a steady 3.3V supply to supply the microcontroller. If you can run everything from 3.3V you don't need to worry about voltage translation, and the risk of damage due to misconnection is minimised

A 3.3V supply might be derived from a 4V lithium cell using a 3.3V low dropout regulator.

Examples of 3.3V devices:

Many nominally 5V LCD modules can run from 3.3V if supplied with a negative contrast voltage, technically an extra supply but the current draw is tiny.

An all 3.3V design avoids the need to translate voltages which will greatly simplify the design process.

Go old school with a single 5V supply

Older generation PIC microcontrollers are designed for maximum performance at 5V. They will operate at lower supply voltages but the maximum recommended clock frequency will be lower at 3.3V.

A 5V supply gives good compatibility with older 5V logic parts such as LSTTL, although some limitations remain. The logic high output of a LSTTL gate may be too low to reliably connect to 5V CMOS without a pull-up resistor.

A 5V supply may help in interfacing with some analogue circuits, since by default interfaces such as ADCs and DACs will have a voltage span of 0-5V when powered from 5V. 3.3V powered devices obviously support a lower 0-3.3V span.

5V circuits may be powered directly from a "USB" 5v power supply, though I believe the USB specification doesn't actually guarantee 5V.

Go dual voltage with both 3.3V and 5V supplies

A dual voltage system means checking device compatibility and implementing level converters where required.

An example of where conversion may not be needed in a dual voltage system is the ENC28J60 Ethernet interface. The inputs that need to be driven by a microcontroller: CS, SCK and SI are 5V tolerant meaning they will not be adversely affected by being driven 1.8V above the device's supply rail. The outputs that need to go to the microcontroller: SO, Clockout and INT will only be driven to 3.3V which is too low for reliable operation of a 5V CMOS input, however if it can be arranged that the microcontroller pins used have "TTL" input thresholds then direct connection is possible.

I've made a list showing logic compatibility and a list of resources for voltage conversion.

Be warned that It is easy to "sleepwalk" into a mixed voltage system, particularly when using development kits, as an example the Arduino Uno R4 is primarily a 5V platform but the WiFi function is provided by a 3.3V powered ESP32 device. The Mikroelektronika "click" platform specifies dual supplies of 3.3V and 5.0V. Many click boards have configuration links for 3.3V or 5.0V compatibility, though the factory default is almost always 3.3V.

Alternatively you may intentionally choose to go dual voltage in order to support a mix of 3.3V and 5V devices running at their specified supply voltages.

Add a -5V, -3.3V or -1.8V supply

A low current negative supply may be inexpensively derived using a charge pump IC such as the ICL7660 or one of its numerous equivalents.

A negative supply may be required for some types of LCD display, or where not actually required it may be convenient, for example where it is desired to use a 5V LCD in a 3.3V design.

A negative supply simplifies processing of analogue signals referenced to ground as some analogue ICs may be powered from +/-5V or +/-3.3V.

The supply range of +/-5V is within the supply range of the 74HC4051, 74HC4052 and 74HC4053 analogue switch ICs.

A 5V ADC or DAC might be powered from supplies of +3.3V and -1.8V, enabling it to process both positive and negative voltages. It is a challenge to convert the resulting logic levels back to 0-3.3V but a series P-FET might be used.

Add +/-15V or +/-12V supplies

+/-15V or +/-12V, often described as simply +V and -V, may be used to power "classic" operational amplifiers such as the 741 and its derivatives, also analogue multipliers such as the AD633. Historically a +/-15V supply allowed for a +/-10V signal range without clipping. More modern devices do not require as much "overhead" and can reliably achieve +/-10V from +/-12V supplies. 

Another use for +/-12V was to power RS232 buffers. The MC1488 quad buffer requires +/-12V supplies (actually the datasheet says +/-9V is sufficient), though it has mostly been superseded by parts with inbuilt charge pumps running from 5V or 3.3V.

For modest current drains of around 50mA it may be worth considering using a DC-DC converter to derive the supplies from 5V.

It is worth checking if you can get away with lower supply voltages, for example +/-5v signals might be handled adequately by "rail to rail" operational amplifiers running from +/-5V.

Run from a 12V or 24V supply

Sometimes it is advantageous to fit a regulator or step-down DC to DC converter to allow a board to accept a 12V or 24V supply. This can go a long way to "foolproof" a design as it is relatively simple to protect against reverse polarity and many step-down converter circuits will accept a wide range of voltages. This may be desirable in systems incorporating DC motors or other high power loads.

A higher input voltage means a lower current. Many broadband routers run from 12V external supplies, even though internally the supply must be converted down. One explanation I have read is that meeting energy efficiency regulations at 5V demands a thicker cable compared to 12V.

An example of multiple supply rails: The 4116 DRAM

Although this was a computer part and was used with 5V microprocessors it actually needed three supplies and ground.

VDD was the device's actual supply rail, normally 12V

VCC was the I/O supply, normally 5V for compatibility with 5V microprocessors

VBB was a low current bias supply, almost certainly going to the device substrate. The normal MOSFET symbol we are all taught has the substrate connected to the source, but in integrated circuits it is common to have the substrate independent. Giving a N-type FET a negative substrate bias makes its threshold voltage more positive and helps to reduce leakage when the device is off. 

The next generation DRAM was the 4164 with four times the capacity. The 4164 only needed a 5V supply. I actually encountered some 1980s computers that could be configured to accept either device, using jumper links to select the appropriate voltages. The reason for the change was a major improvement in NMOS logic technology, though it is possible that the devices may have retained an internal self-generated VBB.