MCP4706: Myths, Wiring, and Debug for Microchip's 8-Bit I²C DAC

September 3, 2026 · AI-generated from the datasheet, fact-checked by two independent LLM critics

The most common misconception about the MCP4706 is that it's a general-purpose DAC you can hang anything on — a waveform generator, a small power driver, a drop-in for any DAC socket. It isn't. The MCP4706 is an 8-bit buffered voltage-output DAC from Microchip with an I²C interface, a nonvolatile EEPROM for power-up defaults, and an output buffer optimized for holding a stable DC level into a load of 5 kΩ or more. Its datasheet's own application examples are all trim-and-calibrate circuits: set points, sensor offset and slope calibration, thresholds, programmable current sources. Treat it as a low-cost digitally-controlled trim element, not an arbitrary-waveform engine, and most of the design decisions below fall into place.

Next step

Start building with MCP4706 → This guide comes from the same grounded, cited datasheet answers — ask the assistant your own MCP4706 questions.

What the MCP4706 actually is

The MCP4706 is the 8-bit member of the MCP47X6 family: a single-channel, rail-to-rail voltage-output DAC with a volatile DAC register, a nonvolatile EEPROM that reloads at power-on, and an I²C serial interface. It runs from a single 2.7–5.5 V supply, drawing 210 µA typ (400 µA max) unloaded, dropping to 0.09 µA typ (2 µA max) in power-down mode (specified with VOUT not connected). The output settles in 6 µs typ within ½ LSB for a quarter-to-three-quarter full-scale step, and the reference can come from VDD or the VREF pin, with an optional 2× gain in the latter case. The datasheet's front-page applications list is Set Point or Offset Trimming, Sensor Calibration, Low-Power Portable Instrumentation, PC Peripherals, Data Acquisition Systems, and Motor Control — and the worked examples behind those are all DC set-and-calibrate circuits, which tells you what the part is really for.

Wiring Up the MCP4706

The external circuit is minimal, but each element has a reason:

Supply decoupling: the datasheet recommends a 0.1 µF ceramic bypass from VDD to ground, plus a 10 µF tantalum in parallel to attenuate high-frequency noise. Both go within 4 mm of the VDD pin. This isn't just good hygiene: when VDD is the reference, noise on VDD is noise on your DAC output, so put VDD/VSS on the analog plane if your board splits supplies.
I²C pull-ups: SCL and SDA are open-drain N-channel drivers — each needs its own pull-up to VDD. The datasheet's guidance is 1–10 kΩ for Standard/Fast mode and <1 kΩ for High-Speed mode. Higher values save power but slow the edges; lower values speed edges at more current. At 400 kHz with 400 pF of bus capacitance, the rise-time limit works out to roughly 885 Ω minimum pull-up, so don't blindly carry 10 kΩ pull-ups onto a fast bus.
No address straps: the slave address is a fixed 4-bit '1100' plus a 3-bit code selected at ordering time (up to 8 devices per bus via order-code suffix). Don't add pull resistors for device selection — pick the right suffix (A0/A1/…) when ordering.
VREF pin: if you run VDD as the reference, the VREF pin is internally disconnected — no external reference, decoupling, or pull-down needed. Only if you select the VREF-pin reference do you need a source there, and if it's unbuffered it must come from a low-impedance source, because the resistor ladder looks like ~210 kΩ with ~29 pF typ.
Output load: the buffer is specified for RL = 5 kΩ to ground with CL = 100 pF, and the datasheet is explicit that load resistance must stay above 5 kΩ for spec-meeting output. For large capacitive loads it prescribes a small series isolation resistor (R_ISO) at the output to preserve stability.

One resolution note: since this is an 8-bit DAC, 1 LSB = VREF/256. A 5 V scale gives 19.5 mV/LSB, while a 2.048 V external reference gives 8.0 mV/LSB — pick the reference that maps the resolution you need onto the span you actually use.

MCP4706 Gotchas to Watch For

The output is not a power driver. The VOUT-vs-load curve collapses hard under light resistive loads — roughly 1.4 V at 100 Ω, recovering to ~4.9 V only at ≥1 kΩ with VDD = 5 V. If your load is lighter than 5 kΩ, buffer it externally; don't hang it off VOUT. The short-circuit limit is 40 mA max source/sink, but that's a fault boundary, not an operating point.

Rail-to-rail still has headroom. Typical endpoints are VOUT(min) = 0.01 V and VOUT(max) = VDD − 0.04 V. Don't design for full-rail output codes.

Absolute accuracy is coarse; repeatability is fine. Offset error is ±0.02% FSR typ (0.75% max), gain error −2 to +2% FSR (typ −0.1%, and per the datasheet's Note 2 this gain error does not include offset error, which is specified separately), and the 8-bit endpoint errors run 0.13 LSB typ / 2.0 max at zero-scale and 0.3 typ / 5.2 max at full-scale. INL is ±0.125 typ / ±0.907 max LSB and DNL ±0.0125 typ / ±0.05 max LSB. Net: don't count on better than ~±2–3 LSB absolute accuracy without calibration, but the DNL is comfortably monotonic-grade — fine for set points where repeatability matters more than absolute value.

Gain settings have conditions. The G bit is ignored (gain = 1) when VDD is the reference. Gain 2× is only valid when the reference comes from the VREF pin AND VREF ≤ VDD/2. An output that looks "too low" is often a gain setting that can't take effect.

The power-up ramp matters. A minimum 1 V/s ramp rate is specified, and ramp rate affects uploading EEPROM contents into the DAC register at power-on. After VDD crosses the ~2.2 V typ POR threshold, wait out the ~60 µs typ reset delay — the I²C interface refuses commands until it elapses.

EEPROM writes are slow and finite. 25 ms typ / 50 ms max per write, 1M-cycle minimum endurance, 200-year retention at 25 °C. Write frequently-changing values to the volatile DAC register and reserve EEPROM for power-up defaults and configuration — never stream updates to EEPROM.

Full-scale steps are slew-limited. Settling is 6 µs typ for mid-range steps, but the slew rate is 0.55 V/µs typ, so a near-5 V rail-to-rail step takes ≈9 µs at the slew limit. A full-scale sweep is limited by slewing, not settling.

I²C speed depends on supply. Standard/Fast mode runs to 400 kHz across the full 2.7–5.5 V range with Cb ≤ 400 pF. Hs-mode at 1.7 MHz (3.4 MHz with Cb ≤ 100 pF) requires VDD ≥ 4.5 V. Also note the inputs are not rail-immune: VIH = 0.7·VDD, VIL = 0.3·VDD — a 3.3 V master driving a 5 V-supplied MCP4706 is marginal at best.

Verifying the MCP4706 Design

Work through these checks at bring-up:

POR and reset delay: right after VDD rises above the POR trip point, send an I²C write — it should be NACK'd until the reset delay elapses. Then poll the status word: the POR status bit must read '1'. If it reads '0', VDD is below VDD(MIN) and the datasheet warns "unreliable device operation should be expected" — check the rail, not the device.
Power-up default: confirm VOUT settles to the EEPROM-stored code, not 0 V (unless you've configured power-down defaults).
EEPROM write integrity: write a known code and config to EEPROM, poll the RDY/BSY bit until it reads '1' (not programming), power-cycle, and verify VOUT matches the stored code. Never issue another EEPROM write while RDY/BSY = '0' — commanding during a programming cycle risks corruption.
Brown-out behavior: ramp VDD down and confirm that on crossing the BOR trip point the device forces PD1:PD0 = '11' (power-down), clears the volatile DAC register to 000h, and clears the VREF and G bits. On recovery, VOUT should return to the EEPROM-latched default, not the last volatile value.
Mid-code accuracy: with VDD as reference and gain = 1, measure VOUT at mid-code and compare against the datasheet's transfer equation (VOUT = VRL × DAC value ÷ 256, ± offset/gain error). A gross mismatch usually points to a wrong VREF/gain configuration rather than a dead part.
Load compliance: verify the actual load is ≥5 kΩ resistive with ≤100 pF capacitance, and re-verify VOUT after any exit from power-down, since the ladder is disconnected during PD modes.
Absolute maximums (boundaries, not test targets): VDD vs VSS −0.6 to +6.5 V; all pins −0.3 to VDD+0.3 V; input/output clamp ±20 mA; SDA/SCL input current 2 mA max; SDA output sink 25 mA; VOUT source/sink 40 mA max; VREF sink 40 µA max; ESD ≥6 kV HBM and ≥400 V MM. Meter the rails for overshoot beyond 6.5 V during hot-plug events.

Known limitation: the exact tPORD reset-delay value and some POR trip-point rows live in timing tables that couldn't be fully read for this guide; the behavioral descriptions above are verbatim from the datasheet, but verify the precise numeric rows against the datasheet's timing tables before finalizing your firmware timeouts.

Troubleshooting the MCP4706

Device NACKs the address. The slave address is a fixed '1100' plus a 3-bit code set at ordering time — a mismatched order-code suffix is the classic "hardware is there but I can't talk to it" failure. Cross-check the suffix against what your master sends. After any Not-Ack, a Start bit must be issued to reset the command state machine; retrying the address without a Start can leave the device stuck. If even the General Call address doesn't ACK, the problem is the bus — pull-ups, wiring, or VDD — not the address.

Bus looks hung. Use the datasheet's Software I²C Interface Reset Sequence (Section 8.9): a second Start bit must be sent before the Stop bit. Why: if the master was reset mid-write and returns with a Start while the device is issuing an ACK, a plain Start+Stop can trigger an erroneous write. The second Start defuses that race.

Need a clean recovery primitive. General Call Reset (second byte 06h) aborts the current operation and reloads EEPROM into the DAC registers — like a POR without the reset delay — and works across multiple devices synchronously. General Call Wake-Up (second byte 09h) forces volatile PD1:PD0 to '00' without touching EEPROM.

VOUT not changing though I²C works. Three usual suspects: (1) PD1:PD0 ≠ '00' — VOUT is high-impedance and the ladder is disconnected from the reference; wake it with a General Call Wake-Up or any write setting PD1:PD0 = '00'. (2) You sent a Write Volatile command (C2:C0 = 00x), which updates only the volatile DAC register — the change won't survive power-up. (3) An EEPROM write is still in progress — check RDY/BSY before issuing another Write All Memory.

VOUT present but wrong voltage. Check the load first: below 5 kΩ the output sags dramatically and the curve collapses at a few hundred ohms. Then check the reference: an unbuffered VREF fed from a high-impedance source droops under the ladder's ~210 kΩ (±20%) load, and a drooping reference looks exactly like a calibration error. Finally, re-check the gain configuration — 2× gain silently does nothing unless the reference is the VREF pin at ≤ VDD/2.

Writes before the device is ready. During the post-POR reset delay the I²C interface will not accept commands; writes get NACKed. Wait out the delay and confirm an ACK before configuring.

Need more resolution or more channels? See the MCP4725 guide for the 12-bit single-channel sibling, or MCP4728 for the 12-bit quad-channel version with an internal reference option — both share this part's EEPROM-backed power-on defaults.

Part page: MCP4706.