Workstations and toolingguide
Read-only SWD and voltage-glitch harness on a locked Nuvoton M483
Bringing up SWD and a MOSFET crowbar glitcher on a live consumer peripheral has a narrower safety margin than bench work on a bare development board. While probing my own Endgame Gear OP1 8k v2 mouse (covered from the USB host side in 8 kHz mouse polling on Windows) to study its locked Nuvoton M483SIDAE microcontroller, three hardware traps surfaced before a single valid fault trial could run: stock OpenOCD adapter initialisation asserted resets and enabled target power, failed SWD transactions echoed ADC voltage words back as fake chip IDs, and connecting a powered target rail to a Faultier crowbar line back-fed the glitcher’s RP2040 through its GPIO ESD diodes until Windows refused to start the USB device with error code 10.
This guide covers the physical bring-up, rail-ownership rules, patched safe-connect OpenOCD driver, and logic-analyzer pulse verification for a read-only SWD and crowbar harness. The companion reference, Nuvoton M483 flash locks and the glitch-oracle trap, covers the Flash Memory Controller (FMC) lock architecture and why a fault-injection sweep on this target fails without an address-sensitive readback oracle.
Constants
Section titled “Constants”Every check in this procedure depends on the target identity, pin map, and sampling constants below.
Target and adapter identifiers
Section titled “Target and adapter identifiers”| Component | Identifier / value | Role in preflight |
|---|---|---|
| Target MCU | Nuvoton M483SIDAE (LQFP-64, Cortex-M4)1 | 512 KB APROM, 4 KB LDROM, no hardware AES or XOM |
Expected SWD DPIDR | 0x2ba01477 | Only accepted target identity check |
| OP1 8k v2 USB (application mode) | 3367:1978 | Normal HID enumeration when powered over USB |
| OP1 8k v2 USB (update mode) | 3367:1977 | Bootloader HID enumeration |
| Nuvoton USB HID ISP IDs | 0416:a316, 0416:3f00 | Mask-ROM / CPU ISP HID signatures (absent on normal boot) |
| Nu-Link Pro USB ID | 0416:511c (product 0x40012009, FW 8062) | SWD debug probe (ICE_DAT, ICE_CLK, /RESET, VCC sense) |
| NuMicro ICP Programming Tool | 3.24.8062r2 | Used once to set Nu-Link Pro I/O voltage to 3.3 V |
| Faultier glitcher ADC clock | 250 MHz / 96 = 2.604 MS/s (~384 ns/sample) | Onboard ADC_CROWBAR 3,000-sample buffer at clock division 1 |
LQFP-64 pin and bottom debug-header map
Section titled “LQFP-64 pin and bottom debug-header map”The M483SIDAE can be probed either on the LQFP-64 legs with spring-loaded SP10 microprobes or on the unpopulated five-hole debug row (H1 through H5, numbered left to right looking at the bottom PCB with the optical sensor lens below the row).
| LQFP-64 pin | Signal | Bottom row hole | Nominal DC voltage | Harness connection |
|---|---|---|---|---|
| Pin 32 | nRESET | H1 | 3.3 V (active-low) | Nu-Link /RESET, Faultier EXT0 trigger, LA D0 (CH0) |
| Pin 33 | PF.0 / SWDIO | H3 | 3.3 V I/O | Nu-Link ICE_DAT (or Faultier Program IO, never both) |
| Pin 34 | PF.1 / SWCLK | H2 | 3.3 V I/O | Nu-Link ICE_CLK (or Faultier Program CLK, never both) |
| Pin 49 | VSS (GND) | H5 | 0 V | Common return for Nu-Link VSS, Faultier Crowbar G, and LA GND |
| Pin 50 | LDO_CAP | None | ~1.26 V | Internal core LDO capacitor node (2.2 uF); leave capacitor installed on VDD route |
| Pin 51 | VDD | H4 | 3.3 V | Target 3.3 V rail; Nu-Link VCC (sense only), Faultier Crowbar D, LA D1 (CH1) |
Architecture overview
Section titled “Architecture overview”The harness separates power delivery, SWD readback, crowbar fault injection, and digital timing verification so that no two tools drive the same net simultaneously.
Component versions
Section titled “Component versions”- Patched xPack OpenOCD built from commit
7bb05570f40dae2c742e6e581db4129468fed6893 with a safe-connect HLA layout patch applied over the OpenOCD-Nuvoton JTAG/SWD driver4. - Nuvoton NuMicro ICP Programming Tool
3.24.8062r(digitally signed by Nuvoton Technology Corporation)2. - Faultier RP2040 crowbar hardware using
ADC_CROWBAR(3,000samples per trigger) and SMA-to-pigtail drain/ground leads. fx2lafw-compatible logic analyzer running undersigrok-cliat1 MHzand2 MHzsample rates.
Step 1: Unpowered continuity and short isolation
Section titled “Step 1: Unpowered continuity and short isolation”Adjacent pins on a 0.5 mm-pitch LQFP-64 package bridge easily under angled microprobe needles, and semiconductor junctions inside the MCU can trigger a multimeter continuity buzzer even when no metallic short exists.
- Disconnect every USB cable (target mouse, Nu-Link Pro, Faultier, and logic analyzer) before switching the multimeter to resistance or continuity mode. Never probe a powered board in resistance or continuity mode: touching pin 51 (
VDD) to ground with a continuity source injects current into the rail and causes an immediate USB reset. - Measure end-to-end resistance along each intended wire path from the tool header to the target pad:
- Nu-Link pin 2 (
VCC) to pin 51 (VDD/H4): under2 ohm. - Nu-Link pin 4 (
ICE_DAT) to pin 33 (SWDIO/H3): under2 ohm. - Nu-Link pin 6 (
ICE_CLK) to pin 34 (SWCLK/H2): under2 ohm. - Nu-Link pin 8 (
/RESET) to pin 32 (nRESET/H1): under2 ohmwhen attached. - Nu-Link pin 10 (
VSS) to pin 49 (GND/H5/ USB shield): under2 ohm.
- Nu-Link pin 2 (
- Measure resistance between adjacent signal nets in both probe polarities:
- Pin 33 (
SWDIO) to pin 34 (SWCLK) must not read near0-2 ohm. - Pin 32 (
nRESET) to pin 33 and pin 34 must not read near0-2 ohm. - Pin 51 (
VDD/H4) to pin 49 (GND/H5) must not read near0-2 ohm.
- Pin 33 (
- If pin 33 and pin 34 read
0 ohm(observed during initial bring-up on 2026-11-13), lift both microprobe needles and measure the board pads and the free probe leads separately. If both are open when separated, the seated needles were bridging the adjacent legs. - Check
SP10header jumpers: eachSP10probe body exposes two header pins tied to the same needle. Use one jumper lead per probe unless intentionally splitting that signal to a passive observer such as a logic analyzer channel.
Step 2: Rail ownership and power-up ordering (avoiding back-powering)
Section titled “Step 2: Rail ownership and power-up ordering (avoiding back-powering)”When a debug probe drives 3.3 V onto the target VDD rail (supply=1) or the target is powered before Faultier USB enumerates, that 3.3 V sits directly on Faultier’s crowbar drain (Crowbar D) and any connected GPIO/sense lines. Current flows from the target VDD net backward through the RP2040 GPIO ESD protection diodes into the glitcher’s internal 3V3 rail. Because the RP2040 core is partially energised through its I/O clamping diodes before its own USB regulator and power-on reset sequence run, the microcontroller enters a wedged state: Windows detects the USB device descriptors (VID:PID appears in Device Manager) but marks the composite device failed with error code 10 (This device cannot start), and usbipd attachment produces no persistent device node.
Enforce three electrical rules across every session:
- Pick one rail owner. Power the target board from its own USB cable (
VBUS4.92 Vfeeding the onboard3.3 Vregulator). Keep the Nu-Link Pro target-power output disabled (supply=0), and leave Faultier’sProgramheader3.3V/VCCpin physically disconnected (front view left-to-right: VCC, CLK, GND, IO, RST). - Keep one SWD master on
SWDIO/SWCLK. Attach either the Nu-Link Pro (ICE_DAT/ICE_CLK) or Faultier’sProgramheader (IO/CLK), never both at once. - Follow strict connect and teardown ordering.
- Connect sequence. (1) Plug Faultier into host USB first and confirm clean USB enumeration with all target leads disconnected; (2) plug in the Nu-Link Pro (with
supply=0) and logic analyzer over USB; (3) wire the ground, SWD, sense, and crowbar leads to the unpowered target; (4) plug in the target mouse USB cable last to energiseVDD. - Teardown sequence. (1) Unplug the target mouse USB cable first so
VDDdrops to0 V; (2) disconnect the harness leads from the target; (3) unplug the Nu-Link Pro and Faultier USB cables last.
- Connect sequence. (1) Plug Faultier into host USB first and confirm clean USB enumeration with all target leads disconnected; (2) plug in the Nu-Link Pro (with
With the target powered over USB and all tools idle, verify DC voltages with the multimeter in DC volts mode:
- Pin 51 (
VDD) to ground:~3.3 V(Nu-Link pin 2 to pin 10 must match within a few millivolts). - Pin 50 (
LDO_CAP) to ground:~1.26 V. Never attachVDD_SENSEto pin 50; a~1.26 Vor~1.8 Vreading on the probe’s voltage channel means the sense wire is onLDO_CAPor floating.
Step 3: Nu-Link 3.3 V configuration and the safe-connect OpenOCD driver
Section titled “Step 3: Nu-Link 3.3 V configuration and the safe-connect OpenOCD driver”OpenOCD can drive a Nu-Link Pro over USB HID, but it cannot configure the probe’s programmable SWD I/O voltage translator, which defaults to 5 V (~4910-4930 mV measured before configuration).
Set 3.3 V I/O once in the official ICP tool
Section titled “Set 3.3 V I/O once in the official ICP tool”- Open the Nuvoton NuMicro ICP Programming Tool (
3.24.8062r)2 with the Nu-Link Pro connected over USB and the target disconnected. - Select M480 Series, click Option next to Start, set Nu-Link Pro IO Voltage = 3.3 V, and click OK.
- Do not click Start, and do not accept any popup offering to unlock or mass-erase the chip (
Target flash is in security lock mode-> cancel immediately). - Click Stop check if the tool is polling for a target, then close the ICP tool completely so it releases the USB HID handle before OpenOCD starts.
Why the stock Nu-Link OpenOCD configuration is unsafe on a locked target
Section titled “Why the stock Nu-Link OpenOCD configuration is unsafe on a locked target”Inspecting OpenNuvoton’s USB JTAG/SWD driver source (src/jtag/drivers/)4 showed that stock adapter initialization performs RESET_HW and RESET_SYSRESETREQ and sends CMD_SET_CONFIG with supply=1 during adapter open. On a target that is already USB-powered and wired to a crowbar harness, supply=1 violates single-rail ownership, and unsolicited reset assertions disturb target state during a read-only check.
To prevent both hazards, apply a dedicated safe-connect HLA layout patch to xPack OpenOCD (7bb05570f40dae2c742e6e581db4129468fed689)3:
- Register a separate safe HLA layout so an unpatched OpenOCD binary fails closed before opening the USB device.
- In the safe connect path, force
if_target_power_supplied(0)(supply=0), skip hardware reset on open, and reject explicitSRSTassertions. - Expose a read-only
hla_command "raw_mdw <address>"backed by the Nu-Link memory-read API so single 32-bit words can be queried without Cortex-MCPUIDtarget examination.
# Inline safe OpenOCD configurationadapter driver hlahla_vid_pid 0x0416 0x511c 0x0416 0x511d 0x0416 0x5200transport select hla_swdadapter speed 200Fail-closed DPIDR check and the voltage-channel echo trap
Section titled “Fail-closed DPIDR check and the voltage-channel echo trap”Do not treat an OpenOCD exit code of 0 as target detection. When the SWD physical layer fails to get a response from the target, the Nu-Link USB driver buffer can retain the preceding voltage-sense response words and print them as the IDCODE:
| Recorded run state | Reported probe voltage | Returned IDCODE / DPIDR | Diagnosis |
|---|---|---|---|
Target disconnected (FW 7246) | 3 mV | 0x00030001 | Invalid: open sense line echo |
Prior to ICP 3.3 V setup | 4910 mV / 4930 mV | 0x0e381022 / 0x0e620ff7 | Invalid: 5 V I/O mismatch |
ICP set to 3.3 V, open SWCLK | 3315 mV | 0x0a4b0bde | Invalid: broken SWCLK contact |
Voltage-buffer echo (06a9, 06a4) | ~1700 mV (floating sense) | 0x06a406a9 | Invalid: raw voltage words echoed as IDCODE |
| Safe layout with open contact | 2963 mV | 0x0b9f0b9f | Invalid: 2963 mV (0x0b9f) echoed twice |
Safe layout after SWCLK repair | 2980 mV (200 kHz) | 0x2ba01477 | Pass: valid Cortex-M4 DPIDR across 3 consecutive runs |
Notice 0x0b9f0b9f in the fifth row: 0x0b9f in decimal is 2975, matching the 2963 mV ADC sense reading. Always require an exact match on 0x2ba01477 before running any memory read or glitch pulse.
Step 4: Qualifying crowbar delivery with logic-analyzer and ADC baselines
Section titled “Step 4: Qualifying crowbar delivery with logic-analyzer and ADC baselines”Once SWD returns 0x2ba01477 and the crowbar is wired to VDD (pin 51) and GND (pin 49), verify that crowbar pulses actually reach the rail before running any timing sweep.
Why onboard ADC misses nanosecond pulses
Section titled “Why onboard ADC misses nanosecond pulses”The glitcher calls configure_adc(ADC_CROWBAR, 3000) before firing glitch(). At the default clock divider of 1, the RP2040 ADC samples at 250 MHz / 96 = 2.604 MS/s, or one sample every ~384 ns. A 4 ns or 16 ns crowbar pulse is 24x to 96x shorter than a single sample period: an unchanged ADC trace at 4 ns proves that the ADC is undersampled, not that the MOSFET failed to fire. Time-resolving nanosecond pulses requires an oscilloscope; without one, use microsecond calibration pulses (1 us to 1 ms) cross-checked by a 1-2 MHz logic analyzer on VDD (D1) and nRESET (D0) to verify physical delivery and reset latency.
LDO_CAP (pin 50) versus VDD (pin 51) calibration
Section titled “LDO_CAP (pin 50) versus VDD (pin 51) calibration”On LDO_CAP (pin 50, ~1.26 V DC with the factory 2.2 uF decoupling capacitor installed), the normalised Faultier ADC baseline sits at 0.388-0.3917 (above FaultierTool selfcheck’s generic 0.15-0.30 open threshold, which assumes an unpowered pin). Pulses of 4 ns, 16 ns, 128 ns (ADC min/max 0.3686/0.4392), and 1 us produced no resolved droop; a 10 us pulse pulled the normalised ADC minimum down to 0.0235.
Moving the crowbar drain to VDD (pin 51, 3.3 V DC, normalised ADC baseline 0.919-0.933) while leaving the LDO_CAP capacitor intact produced clean, repeatable rail pull-downs confirmed simultaneously by the onboard ADC and an fx2lafw logic analyzer:
Programmed VDD pulse | Faultier ADC samples below 0.5 | Logic-analyzer VDD (D1) low width | Logic-analyzer nRESET (D0) response | USB PnP / SWD readback state |
|---|---|---|---|---|
1 us (250 cycles) | Min 0.0392 (3 low samples) | ~1 us (2 low samples at 2 MHz) | High (60,000,000 samples, 0 edges) | Unchanged / locked-response |
10 us (2,500 cycles) | 33 samples (span 488-521) | 27 low samples (at 2 MHz) | High (60,000,000 samples, 0 edges) | Unchanged / locked-response |
25 us | 79 samples | 32-33 us (at 1 MHz) | High | Unchanged / locked-response |
30 us | 95 samples | Measured low (1 MHz) | High | Unchanged / locked-response |
35 us | 111 samples | Measured low (1 MHz) | High | Unchanged / locked-response |
40 us | 127 samples | Measured low (1 MHz) | Low pulse (~35-37 us after VDD fall) | Unchanged / locked-response |
45 us | 143 samples | Measured low (1 MHz) | Low pulse (~35-37 us after VDD fall) | Unchanged / locked-response |
50 us | 158 samples | 62 us (at 1 MHz) | Low pulse (~35-37 us after VDD fall) | Unchanged / locked-response |
100 us | 306-307 samples | 119 us (at 1 MHz) | Low pulse | Unchanged / locked-response |
200 us / 500 us / 1 ms | 637 / 1,741 / 2,983 samples | 247/670/1149 us (at 1 MHz) | Low pulse across all three rungs | Unchanged / locked-response |
Two hardware facts stand out from the ladder:
- External
nRESETassertion starts between35 usand40 usofVDDpull-down, withnRESETfalling35-37 usafter theVDDfalling edge. Shorter pulses (1 usto35 us) droop the rail without triggering an externally visible pin reset. - Every rung disarmed cleanly without dropping USB enumeration, proving that the crowbar path delivers repeatable energy to pin 51 while keeping the peripheral alive.
Verification
Section titled “Verification”Before declaring the harness ready for any readback or fault experiment, confirm every row in this checklist:
| Gate | How to check | Expected pass state | Stop condition |
|---|---|---|---|
| Unpowered isolation | DMM resistance across SWDIO/SWCLK, nRESET, and VDD/GND with all USB unplugged | End-to-end leads under 2 ohm; adjacent nets open | Any adjacent pair near 0-2 ohm |
| Glitcher USB enumeration | Plug Faultier USB in before target power; check Windows Device Manager | Composite device starts cleanly; no warning icon | Windows error code 10 (This device cannot start) -> unplug target power and probe |
| Rail voltage and ownership | DMM DC volts on pin 51 (VDD) and pin 50 (LDO_CAP); OpenOCD safe-layout log | VDD ~3.3 V, LDO_CAP ~1.26 V, if_target_power_supplied(0) | Probe voltage near 5 V or supply=1 |
| Target SWD identity | Run safe-layout ID check at 200 kHz | Exact DPIDR 0x2ba01477 on 3 consecutive runs | Process exits 0 with 0x0b9f0b9f or other voltage-echo word |
| Crowbar delivery | Single 1 us and 40 us VDD calibration pulses with 1-2 MHz LA on D0=nRESET, D1=VDD | 1 us shows VDD dip with nRESET high; 40 us shows nRESET low ~35-37 us after VDD fall | No VDD transition on LA, or board warms up while idle |
Gotchas and lessons learned
Section titled “Gotchas and lessons learned”- Back-powering through GPIO ESD diodes mimics a dead glitcher. When Windows reported code 10 on Faultier’s composite USB device, the hardware was undamaged:
3.3 Von the crowbar drain was back-feeding the RP20403V3rail before USB enumeration. Enforcing a single rail owner (supply=0on Nu-Link,Program VCCdisconnected on Faultier) and plugging Faultier USB in before target power eliminated the fault immediately. - OpenOCD exit status
0is not target detection. Stock scripts that only check whetheropenocdexited cleanly will pass when the target is completely disconnected, because the Nu-Link driver prints voltage-channel words (0x06230625,0x0b9f0b9f) as theIDCODE. Always parse the output log and assertDPIDR == 0x2ba01477. - Repeated
0x2ba01477on memory reads is a lock signature, not memory content. OnceSWCLKcontact was repaired, reading Cortex-MCPUID(0xe000ed00),CONFIG0..CONFIG3(0x00300000),APROM(0x00000000), and peripheral registers viaraw_mdwreturned0x2ba01477at every address. BecauseCPUIDreturned0x2ba01477, OpenOCD decodedPARTNOas0x147and refused normalmdwcommands withTarget not examined yet. Why that happens - and why you cannot run a fault-injection sweep until you have a positive control that beats it - is the subject of Nuvoton M483 flash locks and the glitch-oracle trap. - Never move
VDD_SENSEtoLDO_CAP(pin 50). Pin 50 sits right next to pin 51 (VDD), but it carries the1.26 Vinternal core regulator output. Connecting a probe’sVCCsense pin to pin 50 produces a~1.26-1.8 Vtarget-voltage reading and breaks3.3 Vlevel translation.
File reference
Section titled “File reference”| Artifact or setting | Purpose |
|---|---|
src/jtag/drivers/ (patched xPack OpenOCD 7bb0557) | Adds the safe-connect HLA layout (if_target_power_supplied(0), suppressed connect resets, rejected SRST, and read-only raw_mdw) |
NuMicro ICP Programming Tool (3.24.8062r) option setting | Stores Nu-Link Pro IO Voltage = 3.3 V in the probe before closing the ICP GUI |
openocd.log and manifest.json per SWD check | Records adapter firmware (8062), sense voltage (~2980-3315 mV), clock (200 kHz), and verified DPIDR (0x2ba01477) |
sigrok-cli fx2lafw capture (1 MHz / 2 MHz) | Digital trace of D0 (nRESET) and D1 (VDD) confirming pulse delivery and the 35-37 us reset assertion delay |
Related docs
Section titled “Related docs”- Nuvoton M483 flash locks and the glitch-oracle trap - the companion reference on
SBLOCK/ALOCK/ISPFFlock mechanics, the access-path split between CPU and ICE, and why a repeatable failure signature is not a campaign oracle. - 8 kHz mouse polling on Windows: merged HID reports - host-side USB HID report batching and sensor delta analysis on the same 8 kHz mouse class.
- Agent-assisted reverse engineering - structuring static decompilation and black-box protocol capture before touching hardware pads.
- PCB-as-code verification - programmatic netlist and pad-coordinate verification used before building physical test harnesses.
References
Section titled “References”-
Nuvoton Technology Corp., “NuMicro Family M480 Series Datasheet,” Rev. 3.09, sections 4.2.3 and 8.5.1. https://www.nuvoton.com/export/resource-files/en-us—DS_M480_Series_EN_Rev3.09.pdf ↩
-
Nuvoton Technology Corp., “NuMicro ICP Programming Tool,” version 3.24.8062r. https://www.nuvoton.com/tool-and-software/software-tool/programmer-tool/ ↩ ↩2 ↩3
-
xPack Dev Tools, “xPack OpenOCD,” commit
7bb05570f40dae2c742e6e581db4129468fed689. https://github.com/xpack-dev-tools/openocd ↩ ↩2 -
OpenNuvoton contributors, “OpenOCD-Nuvoton JTAG/SWD drivers,”
src/jtag/drivers/. https://github.com/OpenNuvoton/OpenOCD-Nuvoton/tree/master/src/jtag/drivers ↩ ↩2