A quartz crystal is a passive resonator with two active terminals that only oscillates when a microcontroller, PHY or discrete amplifier drives it, together with two load capacitors sized to the crystal's rated load capacitance (CL). A crystal oscillator (XO) is a powered, self-contained clock source: crystal, amplifier and output driver in one package, needing only a supply voltage and a decoupling capacitor. Choose the crystal when the chip already has an oscillator circuit and you control the layout. Choose the XO when you need guaranteed start-up, one clock for several loads, a frequency above the MCU oscillator's rated range, or a part that a buyer can second-source without a layout review.
The decision looks like an engineering detail, but it lands on the buyer's desk later: a crystal is only interchangeable with another crystal of the same load capacitance, and an XO is only interchangeable with another XO of the same supply voltage, output type and pin-1 function. This guide covers both sides, with a worked example built on parts GlobX currently holds in stock.
Crystal vs oscillator: the comparison chart
The short answer: a crystal is cheaper per unit and draws no supply current of its own, while an XO costs more per unit but removes the load-capacitor design, the start-up risk and most of the cross-reference traps.
| Attribute | Quartz crystal (XTAL) | Crystal oscillator (XO) |
|---|---|---|
| Type | Passive resonator, 2 active terminals | Active module, needs VDD |
| External parts | 2 load capacitors, sometimes a series resistor | 1 decoupling capacitor (typically 100 nF) |
| What sets the frequency accuracy | Crystal tolerance + load capacitor match + PCB stray capacitance | Fixed by the manufacturer, stated as total ppm |
| Typical accuracy tiers | ±10 to ±30 ppm at 25 °C, plus temperature drift | ±20, ±25, ±50 or ±100 ppm all-inclusive |
| Can drive several chips | No, one oscillator circuit only | Yes, within the output's fan-out and trace length |
| Start-up risk | Depends on MCU gain margin and crystal ESR | Guaranteed by the XO datasheet |
| Key cross-reference parameters | Frequency, CL, ESR max, package/pad layout | Frequency, VDD, output (HCMOS/LVDS), pin-1 function, package |
| Best fit | MCU main clock and 32.768 kHz RTC in volume products | FPGA, SoC, Ethernet/PCIe references, multi-load boards, low-volume builds |
One correction to a common belief, which comes up repeatedly in German forum threads such as mikrocontroller.net: an XO is not automatically more accurate than a crystal. A ±10 ppm crystal on a well-matched layout beats a ±50 ppm XO. The XO's advantage is that its ppm figure is guaranteed without any board-level tuning, so the number on the datasheet is the number on the product.
When a bare crystal is the right choice
A crystal is the right choice when the target chip already contains a Pierce oscillator circuit (almost every MCU does) and the product ships in volumes where the crystal's lower unit price and zero standby current matter.
- MCU main clocks. An STM32F103 runs its high-speed external oscillator (HSE) from a 4 to 16 MHz crystal; an 8 MHz crystal plus two capacitors is the standard design, and we cover that MCU's supply situation in our STM32F103C8T6 alternatives guide.
- 32.768 kHz real-time clocks. A tuning-fork crystal draws nothing on its own, which is why battery-backed RTCs almost never use an XO.
- Proven layouts. If the crystal, load capacitors and ground guard have already passed EMC testing on the previous board revision, reusing them is cheaper than requalifying a new clock source.
The condition is gain margin. STMicroelectronics' oscillator design guide (application note AN2867) asks for a gain margin of at least 5 between the MCU's oscillator transconductance and the crystal's critical value. A crystal with a higher ESR than the one the board was designed with eats that margin, and the result is a board that starts at 25 °C on the bench but fails to start cold in the field.
When an oscillator (XO) is the right choice
An XO is the right choice when the clock feeds a chip with no oscillator circuit, when one source must clock several chips, or when the cost of a start-up failure in the field is higher than the XO's price premium.
- FPGAs and SoCs usually take a clock input, not a crystal pair, so an XO or a clock generator is mandatory.
- Several loads. One 25 MHz XO can feed an Ethernet PHY and a switch on short traces; two crystals would need two oscillator circuits.
- Higher frequencies. Fundamental-mode crystals get thin and costly above roughly 50 MHz, and many MCU oscillator cells are only specified up to 25 to 48 MHz. Above that range, designers move to an XO.
- Low-volume and fast-turn builds. Industrial controllers built in hundreds, not millions, rarely justify the engineering hours spent tuning load capacitors and testing cold start. The XO's unit-price premium is smaller than one day of bring-up debugging.
Worked example: load capacitors, and what a crystal swap does to them
The load capacitors on a crystal are calculated from the crystal's rated CL, so swapping in a crystal with a different CL changes the clock frequency even though the part number "looks equivalent". Microchip's application note AN826 gives the standard formula for a Pierce oscillator:
CL = (C1 × C2) / (C1 + C2) + Cstray, with Cstray typically 2 to 5 pF for the PCB traces and the MCU pins.
- The approved crystal is an 8 MHz part with CL = 12 pF. Assume Cstray = 5 pF.
- With equal capacitors, C1 = C2 = 2 × (CL - Cstray) = 2 × (12 - 5) = 14 pF. The nearest E12 value is 15 pF, giving an effective CL of 12.5 pF. Close enough.
- Purchasing sources a "same frequency, same package" crystal with CL = 18 pF because the original is on allocation. The board still has 15 pF capacitors, so the crystal sees 12.5 pF instead of 18 pF. It oscillates above its nominal frequency, typically by tens of ppm for small SMD crystals.
- For an MCU driving a UART, that error is harmless. For an Ethernet PHY or a Bluetooth Low Energy radio with a ±50 ppm total budget, it can eat most of the margin before temperature and aging are counted.
The buyer's rule that follows: when cross-referencing crystals, CL must match exactly, and ESR max must be equal or lower. Frequency and package are necessary but not sufficient. Our cross-reference guide covers the general method; for crystals, CL is the parameter most often missed.
The footprint trap: a 4-pad crystal is not a 4-pin oscillator
A 3.2 x 2.5 mm four-pad crystal and a 3.2 x 2.5 mm four-pin XO share the same outline, but not the same pinout, so one cannot replace the other on an existing board. On a typical 4-pad crystal, pads 1 and 3 are the crystal terminals and pads 2 and 4 are tied to ground. On a typical 4-pin XO, pin 1 is output enable or standby, pin 2 is ground, pin 3 is the clock output and pin 4 is VDD.
Placed on a crystal footprint, the XO's VDD pin lands on a pad the layout grounds, and its output lands on one of the MCU's oscillator pins. The board does not start, and the failure is easily logged as a "bad batch" rather than a wrong substitution. So when an allocated crystal tempts a buyer toward an XO that is in stock in the right package and frequency, the answer on the same PCB is no. Package dimensions alone are never a cross-reference; see the SMD package size chart for why code names and outlines mislead.
XO-to-XO swaps: three parameters buyers miss
Two XOs with the same frequency and package are interchangeable only if supply voltage, output type and pin-1 function also match. The part numbers encode all three, which our part number suffix decoder explains in general terms.
- Supply voltage. A 1.8 V XO on a 3.3 V rail is overstressed; a 3.3 V XO on a 1.8 V rail may not start or may output a swing the receiver cannot read.
- Output type. HCMOS (single-ended), LVDS and LVPECL (differential) are not interchangeable. FPGA transceiver references are usually differential.
- Pin 1: output enable vs standby. With output enable (tri-state), pulling pin 1 low only disables the output; the oscillator keeps running and the clock returns immediately. With standby (power-down), the oscillator itself stops, current drops to microamps, and the clock needs its full start-up time, usually milliseconds, to return. Firmware that toggles pin 1 and expects an instant clock breaks when a standby part replaces an OE part.
A real example from GlobX stock shows all three at once. The IQD LFSPXO056267 (12,000 pcs, date code 21+, active, as of November 2026) is a 24 MHz HCMOS XO for a 1.8 V supply with a standby (power-down) pin 1. It is a valid choice for a 1.8 V, 24 MHz design that wants a power-down mode, and the wrong part for a 3.3 V rail or for firmware that expects tri-state output enable, even though a frequency-and-package search would list it as a match. IQD is a frequency-control manufacturer in the Würth Elektronik eiSos group, which makes it a European source for EU-based OEMs that want to shorten their clock-component supply chain.
Worked example from GlobX stock: one Ethernet board, three clock options
Take an industrial Ethernet node built around the Microchip KSZ8081MNXIA-TR 10/100 PHY (GlobX stock as of November 2026: 2,000 pcs, date code 22+, status active). In MII mode the PHY needs a 25 MHz reference, which can come from a crystal on its XI/XO pins or from an oscillator into XI. There are three ways to supply it:
- 25 MHz crystal on the PHY. Lowest unit cost and the right choice for a high-volume design with a tested layout. The CL and ESR must match the PHY datasheet's crystal requirements, and the ppm budget must hold over temperature.
- A single XO feeding the PHY. Removes the load-capacitor and start-up risk. The XO must be 25 MHz, HCMOS and specified for the rail it sits on, usually 3.3 V on boards of this type, and the PHY datasheet's ppm limit applies to the XO's all-inclusive figure.
- A programmable clock generator. When the board also needs clocks for an MCU, a switch or an FPGA, a clock generator such as the Renesas (formerly IDT) VersaClock 6 5P49V6914A000NLGI8 (3,500 pcs in stock, date code 19+, active, as of November 2026) replaces several XOs with one device running from a single reference crystal, with several configurable outputs. Fewer line items on the BOM also means fewer clock parts to second-source when one goes on allocation.
The right option depends on volume and how many clocks the board needs. One clock at high volume points to the crystal; one clock at low volume points to the XO; three or more clocks point to a generator, which also helps the BOM cost reduction effort.
A 5-question decision checklist
Answer these five questions in order; the first "yes" decides the clock source.
- Does the receiving chip lack an oscillator circuit (FPGA, many SoCs, clock buffers)? Use an XO or clock generator.
- Do three or more chips need clocks at different frequencies? Use a clock generator.
- Is the frequency above the MCU oscillator cell's rated range, or above roughly 50 MHz? Use an XO.
- Is the build volume low, or is a field start-up failure expensive (remote metering, safety-related equipment)? Use an XO.
- None of the above? Use a crystal, and record CL, ESR max and pad layout on the approved vendor list so that later substitutions are checked against them.
For the second-source step itself, the second-sourcing guide covers how to qualify an alternate without a full requalification.
Need crystals, oscillators or clock generators?
Send us the part number and the parameters that matter (frequency, CL or VDD, output type, pin-1 function, package) and we will quote stock that matches all of them, not just the frequency. Browse current standard clock oscillators or request a quote for an allocated crystal; GlobX answers within 24 hours from Germany.