Designing Basic I2C Bus Expansion Networks Using Hardware Multiplexer Components
Hardware I2C multiplexers isolate identical peripheral addresses and split capacitive loads across independent, switchable bus branches.

Topology
Address collisions break standard two-wire designs as soon as multiple sensors share a hardwired factory address like 0x68. An analog bus multiplexer splits a single primary controller trunk into separate downstream branches, allowing targeted channel selection while keeping the physical capacitive load of inactive segments off the main line.
Hardware multiplexers such as the TCA9548A, PCA9546A, and PCA9544A switch downstream devices using bidirectional pass-gate field-effect transistors placed between the bus master and targets. Each channel requires its own pull-up resistors to its local rail, allowing downstream segments to run at distinct voltage domains between 1.65 volts and 5.5 volts. The controller selects an active sub-bus by sending a single command byte to the multiplexer internal control register at its base address, which is set between 0x70 and 0x77 using external address pins A0, A1, and A2.
Channel isolation preserves signal integrity by preventing unselected branch capacitance from loading the active upstream controller.
Topologies generally fall into tree, star, or daisy-chained layouts depending on cable runs, enclosure boundaries, and cumulative bus capacitance. In an eight-channel star layout, the multiplexer sits near the center of the main controller board, running short, controlled-impedance traces out to local sensors. Remote sensor drops add distributed stray capacitance, typically 30 to 80 picofarads per meter on unshielded twisted-pair wiring.
When a pass gate turns on, its internal resistance ~ usually 4 ohms to 30 ohms based on supply voltage ~ combines with branch capacitance to form an RC filter that alters signal rise times.

Routing Downstream Signal Paths
Routing traces across dense circuit boards requires clear separation between serial clock and serial data lines to prevent capacitive cross-coupling. Under the UM10204 specification, total bus capacitance is capped at 400 picofarads for Standard-mode (100 kilobits per second) and Fast-mode (400 kilobits per second), expanding to 550 picofarads for Fast-mode Plus (1 megabit per second).
Unselected channels sit high-impedance, keeping downstream capacitance isolated from the host. This separation lets a single controller drive twenty identical sensors split across four branches without exceeding bus capacitance limits. Bypass capacitors belong directly next to the multiplexer power pins: combining a 0.1 microfarad ceramic capacitor with a 10 microfarad tantalum or low-ESR ceramic capacitor absorbs switching transients when channels engage.
Downstream branches benefit from symmetric clock and data trace pairs routed over a continuous ground reference plane. Adding ground flood fills between clock and data traces cuts down crosstalk during Fast-mode edge transitions, where slew rates reach 300 nanoseconds or faster. Because pass-gate on-resistance rises noticeably at supply voltages under 2.5 volts, selecting the rail voltage early is essential when designing the channel topology.

Silicon
Pass-gate multiplexers differ in internal architecture, propagation behavior, and fault isolation mechanisms. Selecting the right component involves weighing internal on-resistance, input-to-output propagation delay, interrupt routing logic, and power-on reset behavior across the operating temperature range.
The PCA family represents legacy silicon built with standard N-channel metal-oxide-semiconductor pass-gate arrays. The TCA series updates this with lower-voltage core logic that runs down to 1.65 volts, compared to the 2.3-volt floor of PCA parts. The table below outlines core electrical and functional parameters for common industrial I2C multiplexers and switches.
| Device Part Number | Downstream Channels | Supply Range (V) | Maximum Ron (Ohms) | Propagation Delay (ns) | Hardware Reset Pin | Active Interrupt Inputs |
|---|---|---|---|---|---|---|
| TCA9548A | 8 | 1.65 to 5.5 | 25 | 0.3 | Yes | No |
| PCA9548A | 8 | 2.3 to 5.5 | 40 | 0.4 | Yes | No |
| PCA9546A | 4 | 2.3 to 5.5 | 35 | 0.3 | Yes | No |
| PCA9544A | 4 | 2.3 to 5.5 | 35 | 0.3 | No | Yes (4) |
| TCA9543A | 2 | 1.65 to 5.5 | 15 | 0.25 | Yes | Yes (2) |
| PCA9540B | 2 | 2.3 to 5.5 | 30 | 0.3 | No | No |
Low on-resistance keeps output low levels from drifting above receiver input thresholds. When a downstream sensor pulls the data line low, it has to sink current through the internal pass transistor from both its local pull-up and the main host pull-up.

Voltage Level Translation Dynamics
Internal N-channel pass transistors naturally act as open-drain level translators when upstream and downstream pull-up voltages differ. Powering the multiplexer supply rail at the lower of the two bus voltages clamps signal swings on the lower-voltage side while handling the higher pull-up voltage on the other side.
A host running at 3.3 volts can talk to a 1.8-volt sensor bank using a multiplexer powered at 1.8 volts. Gate clamping prevents overvoltage stress on sensitive low-voltage input buffers. Pass transistor threshold voltages ~ typically 0.7 volts to 1.1 volts across process corners ~ determine the exact conduction cutoff point on rising edges.
From there, the downstream pull-up resistor pulls the signal the rest of the way up to its local rail.
Standby current matters in low-power systems. Modern TCA parts draw under 3 microamperes in standby, whereas older PCA dies draw up to 10 microamperes at temperatures near 85 degrees Celsius. In battery-powered hardware, leaving unselected channels active can cause continuous leakage through external pull-ups if downstream lines float at intermediate levels.

Pullups
Segmented multi-channel buses require dedicated resistor calculations for each branch. Because pass transistors place upstream and downstream pull-ups in parallel when active, single-bus pull-up formulas will undersize resistance values on switched topologies.
When channel 0 is selected, the equivalent pull-up resistance seen by a sinking device is the parallel combination of the upstream and branch pull-ups, plus the multiplexer on-resistance. Standard I2C drivers are rated to sink 3 milliamperes while holding an output low voltage (Vol) of 0.4 volts or less. If equivalent resistance drops too low, sink current exceeds 3 milliamperes and pushes Vol above the maximum input low threshold (Vil) of 0.3 times Vdd, leading to corrupted data or bus hangs.
A sinking peripheral driver must absorb current from both upstream and downstream pull-up networks simultaneously across an enabled channel.
Maximum pull-up resistance depends on total branch capacitance and target rise time (Tr). Fast-mode rules require rise times under 300 nanoseconds from 0.3 Vdd to 0.7 Vdd, governed by Tr = 0.8473 × Rpullup × Ctotal. Minimum pull-up resistance is set by the current sink limit: (Vdd minus Vol) divided by 3 milliamperes.
Take a 3.3-volt branch with 180 picofarads of lumped load across trace routing, pin inputs, and sensor stubs. To keep rise time under 300 nanoseconds, maximum branch resistance is 300 nanoseconds divided by (0.8473 × 180 picofarads), which works out to 1,967 ohms. Upstream pull-up resistance is fixed at 4,700 ohms on the host controller board.
Placing a standard 3,300-ohm resistor on the branch gives a parallel resistance with the 4,700-ohm master pull-up of about 1,937 ohms. Adding 25 ohms of pass-gate on-resistance brings effective sinking resistance to 1,962 ohms. At 0.4 volts Vol, sink current equals (3.3 volts minus 0.4 volts) / 1,962 ohms, or 1.48 milliamperes.
This sits safely under the 3.0-milliampere driver limit while meeting Fast-mode timing requirements.
Using 1,000-ohm pull-ups on both ends drops effective resistance below 500 ohms, drawing over 5.8 milliamperes during low states. Under that load, weak drivers struggle to pull the line below 0.6 volts, triggering framing errors. Treating each branch as an isolated capacitive cell when checking parallel sink currents prevents these failures.

Timing
Firmware timing routines must accommodate arbitration, control register writes, and line settling times during multiplexer operations. Switching channels triggers transient electrical behavior that requires intentional sequencing before addressing downstream devices.
A channel update starts when the controller generates a start condition, sends the 7-bit multiplexer address, sets the write bit (0), waits for acknowledge, and writes the data byte representing the new channel mask. Switching occurs on the rising edge of the acknowledge clock pulse. At that moment, host clock and data lines see quick capacitive transients as uncharged downstream copper and input pins attach to the bus.
While silicon propagation delay is minimal at roughly 0.3 nanoseconds, switched branch settling time is the real timing constraint. If downstream capacitance charges from ground or bleeds off through a partially open gate, immediate reads can sample dirty line states. Adding a 2 to 5 microsecond dead-time delay in firmware right after writing the channel register prevents corrupted start sequences.
Bus speeds up to 1 megabit per second enforce tight setup and hold windows for Fast-mode Plus. Clock low time needs at least 500 nanoseconds, and clock high time needs at least 260 nanoseconds. Pass-gate series resistance slightly slows down falling-edge RC constants, though fall times mostly depend on driver pull-down strength.
The following routine avoids timing errors when switching channels:
- Deactivate current branches by writing 0x00 to the multiplexer control register, disconnecting all downstream lines from the main trunk.
- Execute a settling wait in firmware lasting a minimum of three microseconds to allow residual line charges across disconnected nodes to stabilize.
- Transmit channel selection byte targeting the desired branch bit in the control register, observing standard I2C acknowledge response times.
- Allow bus recovery duration before initiating the subsequent start condition, ensuring downstream pull-up networks restore lines to the idle rail state.
- Issue peripheral start frame addressing the target sensor on the newly isolated, stable physical channel segment.
Skipping these delays causes sensitive peripherals to register false start or stop conditions. In multi-master systems sharing a multiplexer, bus free time between stop and start conditions (tBUF) needs to expand past the standard 1.3-microsecond Fast-mode floor to give disconnected branch capacitance time to discharge.

Faults
Industrial cabling exposes I2C lines to interference and physical failures that can lock up a bus. A single peripheral holding its data line low will crash the primary controller trunk unless the multiplexer can isolate the fault.
Stuck-low states happen when a peripheral state machine freezes during a read, waiting for clock pulses that never arrive while holding SDA low. Connected straight to the host, one frozen sensor takes down the entire system. A multiplexer contains the damage: writing to the control register to deselect that channel physically cuts off the shorted branch, freeing up the rest of the bus.
Deselecting a failed branch disconnects the stuck low line, immediately liberating the primary bus controller trunk.
If SDA gets held low on the main upstream bus, the host cannot send standard I2C stop conditions to the control register. Hardware designs fix this by wiring an active-low reset pin (/RESET) directly from the microcontroller to the multiplexer. Pulsing /RESET low for at least 6 nanoseconds forces the multiplexer back to its power-on state, opening all pass gates and releasing the host bus.

Can Firmware Routine Recover Frozen Downstream Nodes?
Isolating a broken branch keeps the rest of the board running, but recovering the stuck sensor usually requires clearing the fault without toggling global power. Host recovery firmware can run a clock-clearing routine to unstick the peripheral before marking the channel as faulted.
The host isolates all other branches, connects to the affected channel, and flips its SCL pin to a standard push-pull output. Firmware then toggles SCL up to nine times at roughly 50 kilohertz, checking SDA between cycles. The stuck peripheral sees these edges as clock pulses and steps through its internal register until it releases SDA back to high impedance.
The host then sends a valid stop condition, reinitializes the multiplexer, and resumes normal operations.
ESD strikes on external wiring harnesses cause another class of line faults. Internal clamping diodes protect multiplexer pins up to 2 kilovolts under the human body model, but strong voltage spikes can trigger silicon latch-up, pulling heavy supply current. Adding low-capacitance TVS arrays (rated under 5 picofarads) directly at terminal block inputs keeps ESD spikes from latching up the silicon and pulling downstream lines to ground.

Layout
Moving from an I2C multiplexer schematic to a physical board layout requires clean trace routing, careful component placement, and consistent stackup design. Fine-pitch packages introduce parasitic coupling that degrades noise margins over longer bus runs.
Multiplexers like the TCA9548A come in TSSOP-24 packages (0.65 mm pin pitch) and WQFN-24 packages (0.5 mm pitch). The 4.0 mm × 4.0 mm WQFN form factor saves significant space, but requires tight assembly tolerances to prevent solder bridging across channel outputs. Traces should fan out from fine-pitch pins at 0.127 mm widths before widening to standard 0.2 mm traces to minimize series resistance.
Stackup choices directly affect capacitance. Signal lines routed over continuous ground planes have impedances set by trace width, dielectric thickness, and board materials. To control crosstalk between adjacent clock and data traces, space them apart by at least three times the distance between the signal layer and its reference plane.
Recommended physical constraints include:
- Package thermal pads require solder mask defined boundaries with an array of four 0.3-millimeter diameter thermal vias connected to internal ground planes.
- Bypass capacitor traces connect directly to IC supply pads without intervening via transitions to eliminate parasitic trace inductance.
- Channel trace routing enforces equal trace lengths across matching clock and data pairs within a tolerance window of 2.5 millimeters.
- External pull-up placement positions termination resistor arrays immediately adjacent to multiplexer channel pins to prevent unterminated routing stubs.
Running branch traces through multi-pin connectors adds 2 to 7 picofarads of stray capacitance per pin. Avoid placing active switched bus channels on adjacent connector pins; dropping ground pins between channel pairs helps block inductive and capacitive coupling. On four-layer boards, high-speed signal lines must never cross split power planes beneath multiplexer routing zones, as return path breaks alter edge shapes and shift logic thresholds.

Procurement
Sourcing hardware multiplexers for production means balancing electrical specs against supply chain stability, component lifecycles, and board costs. Procurement teams look at package options, pin-compatible alternatives, and minimum order quantities to keep lines running smoothly.
Standard components like 9548 and 9546 switches have drop-in alternatives across major semiconductor suppliers, but prices vary by package and temperature rating. SOIC-24 packages yield higher assembly pass rates on basic SMT lines, but carry a 15 to 25 percent cost premium over TSSOP and WQFN options due to package size. Sourcing engineers balance target assembly yield against unit costs when choosing packages for automated placement.
Automotive-qualified parts (AEC-Q100 Grade 1 or Grade 2) often carry lead times of 16 to 26 weeks during supply crunches, compared to 8 to 12 weeks for industrial equivalents. Teams should monitor product change notifications (PCNs) for silent silicon revisions that adjust input hysteresis or ESD protection without changing base specs. Contracts without explicit PCN requirements leave production vulnerable to unexpected timing shifts when updated silicon hits the floor.


