Boundary-Scan/JTAG: Tackling High Power Density and Thermal Management Challenges in Power Supply and Cooling System PCBs

Boundary scan checks accessible control nets on power and cooling PCBs; loaded measurements verify conversion loss, fan response and thermal performance.

A power board can pass a JTAG interconnect test and still overheat under load. Boundary scan helps find accessible digital connection faults before enabling a high-energy power stage. Conversion efficiency, switching waveforms, fan operation and cooling performance need separate measurements under defined operating conditions.

For power supply and cooling system PCBs, the useful decision is where to place the handoff between those tests. Establish the control connections first, confirm that telemetry is valid, then apply an approved load and measure the response. For the broader choice of electrical test methods, see boundary scan, flying probe and functional test selection. The power-specific handoff depends on supply domains, actuator states and the measurements available on the actual board.

What can JTAG verify on a power or cooling controller?

A JTAG connector does not identify all the tests a device supports. Ordinary boundary scan exercises supported boundary cells around digital pins. A processor debug interface, an FPGA monitoring block and an embedded test program may share access hardware while providing different functions.

Use the exact device and package BSDL files, schematic and assembled netlist to establish which nets are controllable and observable. Some ports are not scannable; power pins and analog feedback nodes must not be counted as tested merely because the package has a TAP. TI's BSDL documentation distinguishes scannable ports from non-scannable linkage ports.

Question to answer Suitable access or measurement What a passing result does not establish
Is an accessible enable, status or digital control net connected correctly? Approved boundary-scan vectors, with suitable cells and electrical states at the relevant endpoints Correct switching behavior, analog thresholds or protection timing
Can a controller read and write a fan or power-management register? A real SMBus/PMBus master, test firmware or a documented bridge Accurate sensor calibration, power-stage efficiency or correct physical response
What is the FPGA reporting for die temperature or an internal rail? The device's documented monitoring instrument and access configuration The temperature of every component, PCB surface or coolant path
Does the converter maintain its specified output under load? Powered functional test with measured input, output and load conditions Full environmental qualification or long-term reliability
Does the cooling system respond to a fault? Controlled fan/sensor fault stimulus and time-correlated electrical and thermal measurements Coverage of every field fault or an automatic safety certification

A PMBus controller may be reached by a processor that also has JTAG, but that does not make PMBus a JTAG instruction. The test plan must identify the actual bus master, firmware or bridge, supported commands and recovery behavior. Programming support and secure-boot verification also require their own device-specific procedures.

How should testing progress from logic power to a loaded converter?

“Before power-up” needs a precise meaning. Scan devices and the required I/O domains normally need their specified supplies. It may be possible to test control connections while the main power stage remains disabled; it is not a completely unpowered test.

Define the fixture's independent means of inhibiting the power stage before loading vectors. A scan update must not unexpectedly enable a MOSFET driver, pump or fan. Check power-domain sequencing and the risk of driving an unpowered receiver through its protection structures. The approved procedure should state what happens if the test controller disconnects, the firmware stops or a supply disappears.

Stage Entry condition Evidence required before proceeding
Unpowered assembly checks Stored energy discharged; board and fixture identified Inspection and specified electrical checks, correct assembly variant and no unresolved fault that prevents energizing
Control-domain power and scan Required logic rails valid; power-stage inhibit independently established Chain identification, supported instructions, covered-net results and explicit exclusions
Register and sensor checks Documented interface available; configuration and ownership known Readback, sensor identity, valid data and defined handling of stale, missing or out-of-range results
Limited-load operation Approved load, supply limits and shutdown conditions Output voltage, current, relevant waveforms and commanded-versus-measured fan response
Rated-condition and fault validation Earlier gates accepted; qualified setup and personnel Approved load/ambient cases, temperature locations, protection response, recovery and pass/fail limits

Static control-net testing does not prove PWM amplitude, edge quality, dead time, propagation delay or ringing at the gate driver. Measure the applicable waveforms at the required operating points. Similarly, a digital power-good indication is useful evidence, but it does not replace a voltage measurement when the acceptance criterion concerns regulation or transient droop.

For a high-current design, include copper distribution, connections and thermal paths in the manufacturing review. A heavy-copper PCB may be appropriate, but copper weight alone does not establish acceptable voltage drop or temperature rise. Release the fabrication drawing and stackup together with the electrical and thermal requirements.

Can JTAG provide useful temperature and voltage readings?

Yes, when the selected device exposes a documented instrument. AMD's UG480 XADC overview, version 1.11, describes internal supply-voltage and die-temperature monitoring in 7 series FPGAs and Zynq-7000 SoCs. Those conversion results can be accessed through the TAP; the on-chip monitoring path does not require a user-instantiated XADC merely to read it through JTAG.

This uses a dedicated XADC DRP access mechanism, rather than ordinary EXTEST measuring analog values. The UG480 DRP JTAG interface chapter also documents access restrictions: the bitstream can limit or disable JTAG access, and use of the dedicated Zynq PS-to-XADC interface disables that external JTAG path. Read transactions involve arbitration; requesting the result too early can return old data.

Record the device, configuration, measurement channel and timestamp with the result. Keep ADC sampling, register refresh and host logging intervals distinct. A fast converter does not imply an equally fast, synchronized stream through the test software.

A die sensor, a remote diode, a board-mounted sensor and a thermal camera observe different locations. Correlate their readings under a defined workload and cooling setup. Surface emissivity, sensor attachment, airflow and thermal contact can change the interpretation. Do not substitute an accessible FPGA temperature for an unmeasured power transistor junction temperature.

How do you verify fan speed rather than just a PWM command?

A written duty-cycle register proves a command was accepted, not that a fan is turning or moving enough air. Check the installed fan's supply, PWM electrical requirements, tachometer output and pulses per revolution. Then compare the command with measured response and the cooling result.

The TI AMC6821 offers a useful, specific example. It is an SMBus/I2C temperature monitor and fan controller, not a JTAG device. A production fixture or approved test firmware can access it through that bus; a JTAG-based tool needs an implemented route to do so.

In the AMC6821 datasheet, SBAS386C, revised July 2007, pages 17–18 describe a 100 kHz internal clock used to measure the tach period. With monitoring enabled, the pulses-per-revolution setting matched to the fan and a valid, nonzero count:

RPM = (100,000 × 60) / TACH-DATA

For an illustrative TACH-DATA value of 2000, the result is 3000 RPM. The configured measurement already accounts for the supported two or four pulses per revolution; do not divide the result by that number again.

AMC6821 check Why it matters to the test
Enable monitoring and match the fan's pulses per revolution A plausible number with the wrong configuration is not a valid speed result
Read the low byte before the high byte Reading the low byte freezes the high byte until the pair has been read, avoiding a mixed result
Account for the selected update interval TACH-FAST selects a 250 ms or 1 s update interval; faster host polling does not create fresh measurements
Check the operating mode and data validity Some low-duty-cycle configurations stop updating the tach measurement; an old nonzero reading does not prove continued rotation
Apply the approved stall and recovery test A normal RPM result alone does not verify the alarm, host response, restart behavior or cooling effectiveness

Use the complete datasheet configuration, including required register bits, rather than treating this example as an initialization sequence. Handle zero counts, overflow, absent pulses and communication faults separately. Measure fault-response timing from a defined stimulus to the required output; the tach update interval is only one contributor.

How much heat does a high-current conversion path add?

Consider a deliberately hypothetical steady-state conversion chain from a 48 V input to a 1 V, 100 A load. Output power is 100 W. Assume the efficiency is measured between those endpoints and includes all intermediate conversion stages in the stated boundary. This does not prescribe direct, single-stage 48 V-to-1 V conversion.

Conversion loss = output power × (1 / efficiency − 1)

Assumed full-chain efficiency Output power Required input power Conversion loss
90% 100 W 111.11 W 11.11 W
95% 100 W 105.26 W 5.26 W

The approximately 5.85 W difference is conversion loss at the same output, not the entire system's heat reduction. The 100 W delivered to the load still has to be accounted for. State whether auxiliary supplies, fans, connectors and distribution losses are inside or outside the measured efficiency boundary.

JTAG does not measure these losses simply by reading a status bit. Use suitable input/output measurements and account for instrument uncertainty, particularly when subtracting two similar power values. Record the load, input voltage, cooling arrangement and stabilization criterion. The same calculated loss does not imply the same junction temperature in different layouts or enclosures.

A processor workload can supply a useful thermal stimulus when its software and hardware behavior are controlled and measured. It is not interchangeable with an electronic load: power management, throttling and workload activity can change the current profile. An arbitrary JTAG port does not provide a standard command to activate a particular tensor core or generate a calibrated current step.

What coverage and release evidence should the supplier quote?

HILPCB provides DFM/DFT consultation, including JTAG-chain and test-clock advice, plus test-program and workflow development. Integrated services can combine SPI, AOI, X-ray, flying probe, JTAG and FCT; engineering also supports heavy copper, high-Tg materials, thermal paths, heatsinks and embedded copper coins. Confirm fixture ownership, access, program version, powered states and report coverage in the order.

Ask for coverage by defect and circuit, with the denominator defined. A percentage of accessible digital nets is not a percentage of all assembly faults, and neither establishes power or thermal qualification.

In particular, a rail can behave normally even if one of several parallel decoupling capacitors is open or missing. Ordinary boundary scan cannot confirm each capacitor's individual connection, value or ESR from the rail's logic state. Assign appropriate component/assembly inspection and electrical methods, and validate the PDN under its specified conditions. Keep residual coverage gaps visible rather than counting the whole capacitor bank as proven.

For an RFQ, submit the schematic, netlist, BOM and exact packages; BSDL and configuration files; rail sequence and inhibit states; fan/controller part numbers; connector pinout; and the approved load and thermal test requirements. Include fixture access and any coating or enclosure restrictions. Protected debug access may require a controlled manufacturing state and an approved final lock procedure.

Request these deliverables explicitly:

  • A test-coverage report tied to the netlist, BSDL, test program and board revisions, with untested nets and components identified.
  • The power-up, power-down and abort sequence, including independent energy controls and the outputs left active after a failure.
  • Register configuration, data-validity rules, calibration references and time-stamped measurements for the specified sensor and fan checks.
  • Input/output load results and thermal records with locations, ambient conditions, airflow or coolant setup, dwell/stability criteria and acceptance limits.
  • Failure disposition, rework and retest rules, plus the report format and traceability fields agreed for the build.

For turnkey PCB assembly, have HILPCB confirm the available test route, customer-supplied fixtures or programs, engineering work and report scope against the submitted package. The product owner retains responsibility for the system operating envelope and qualification plan; the manufacturing quotation should identify which board-level evidence will actually be delivered.

Frequently asked questions

Can boundary scan test a power board before it is powered?

The scan device and its required supply domains normally need power. An approved fixture may keep the main power stage disabled while powering the control circuitry for structural tests. Completely unpowered checks and powered boundary scan are different steps.

Does a passing JTAG test prove every decoupling capacitor is connected?

No. Parallel components can mask an individual open connection, and ordinary boundary cells do not measure every capacitor's value or ESR. Specify suitable inspection and electrical coverage, then validate the power-distribution network under the required operating conditions.

Can JTAG read the temperature of any CPU or FPGA?

No. Access depends on the device's documented instruments, configuration and permissions. AMD's 7 series and Zynq-7000 XADC is a specific example of monitoring accessible through a dedicated JTAG path. It does not establish a universal CPU interface or measure the whole PCB temperature field.

Why can a fan report RPM while cooling is inadequate?

The reading may be stale or incorrectly configured, and even a valid speed does not prove airflow through the intended path. Check measurement validity, fan direction and installation, obstructions, thermal contact and temperature response under the approved load.

Can a scan script generate a known thermal load?

Only when the particular device, test logic or firmware supports the requested operation and the resulting power is measured. Generic boundary scan does not provide calibrated CPU workload control. Use a defined workload or load instrument and record the electrical and thermal conditions.

What should a power-board test report contain besides pass or fail?

It should identify the board and test revisions, coverage and exclusions, power states, sensor configuration, valid measurements, load and cooling conditions, acceptance limits and failure handling. Match the report to the quoted manufacturing scope and the product owner's qualification requirements.