Failure analysis · Package inspection

Find the fault
before you cut.

THz near-field probing adds a contactless, non-destructive localization step ahead of destructive preparation — electrically, by time-domain reflectometry with a 0.6 ps pulse launched at the device, and structurally, by near-field imaging through mold compound and polymers.

Send us a sample Two routes into the package
THz time-domain reflectometry — pulse launch to fault location Schematic
01 Launch at the probe tip 02 Propagate along the line 03 Read the echo 04 Locate & classify fs gate · 780 nm TDR probe Tx Rx on-chip line · Cu-CPW 0.6 ps rise · launched at the DUT no galvanic contact short open end incident pulse echo 0 1 mm 2 mm vg ≈ 121 µm/ps · α ≈ 0.64 dB/mm probe-internal short · ρ < 0 open · ρ > 0 echo polarity → defect type TIME — 0…45 ps echo delay → distance x = vg · t / 2 distance to fault from echo delay 0 1.0 2.0 mm 1.00 mm SHORT ρ < 0 2.00 mm OPEN ρ > 0 two-event res. ≈ vg·tr/2 ≈ 36 µm partial (semi) shorts resolved
0.6 ps
Pulse rise time at the DUT
≈ 35–40 µm
Two-event resolution (Cu-CPW)
Open / short
Classified from echo polarity
Contactless
Field coupling, no galvanic contact
The problem

Localize first, cut second.

Failure analysis is a chain of expensive, destructive steps. The more precisely a fault is localized before the first cut, the less preparation, FIB time and sample material it costs — and the less likely the defect is destroyed on the way.

  1. 01 · Non-destructive

    Failure localization

    LIT, TDR, X-ray …

    New THz TDR · THz near-field imaging

  2. 02 · Destructive

    Target preparation

    Laser, FIB, RIE, wet chemistry, milling, abrasive blasting …

  3. 03 · Imaging & physical analysis

    Failure analysis

    SEM / EBSD, EDX, TEM, ToF-SIMS …

  4. 04 · Material

    Characterization

    Image correlation, ESPI …

Steps 02–04 are where time and cost accumulate. Every gain in localization accuracy at step 01 shrinks the volume that has to be prepared.

Why rise time decides resolution.

In TDR, two discontinuities can only be told apart if they are further apart than roughly vg·tr/2. Electronic TDR is limited by its microwave signal path — the THz pulse is generated and detected on the probe itself, directly at the device.

TDR approach Signal path to the DUT Rise time at DUT Rise time, linear scale
All-electronic TDR Pulse generator, cables, microwave probe > 10 ps
Electro-optic TDR, microwave chain On-chip photoconductive switch, waveguide transition, hollow waveguide, microprobe (< 110 GHz) ≈ 6 ps*
THz TDR microprobe None — photoconductive switches integrated in the probe tip 0.6 ps

* Datasheet value of a commercial electro-optic TDR system. THz figure: probe-integrated pulse without DUT, 0.6 ps rise time, 2.0 ps FWHM (Nagel et al., Opt. Express 2011).

The method

Two routes, one probe platform.

THz penetrates mold compound, polymers, ceramics, glass and high-resistivity silicon; metals reflect. The radiation is non-ionizing and needs no sample preparation.

Electrical route

THz time-domain reflectometry

A sub-picosecond pulse is launched into the interconnect directly from the probe. Every discontinuity reflects part of it; the echoes locate and classify faults along the line — open, short or partial.

  • Probe placed at a launch point on the line
  • Distance to fault from the echo delay
  • Defect type from the echo polarity
Structural route

THz near-field imaging

The probe is raster-scanned over the package in transmission or reflection. The result is a set of amplitude and phase maps — and in reflection, time-of-flight slices through the package depth.

  • Contrast for dielectric and polymer structures and their interfaces
  • Complementary to X-ray
  • Depth information in reflection mode

How the probe couples to the device.

Step 01 · Gate

A switch that is open for < 1 ps

An LT-GaAs photoconductive switch sits in a µm-scale electrode gap at the very tip of the probe. A femtosecond gate pulse — 780 nm, or 1550 nm-compatible versions — makes the gap conductive for less than a picosecond.

Step 02 · Couple

Pure field coupling, no contact

The local near field of the device biases the gap. There is no galvanic contact and no RF interconnect; the stand-off ranges from a few µm to about 100 µm. The unscreened, non-resonant tip keeps the field perturbation low.

Step 03 · Sample

Amplitude and phase of E(t)

The gate-synchronous photocurrent is proportional to the field. A delay scan reconstructs the full waveform. Probe orientation selects Ex, Ey or Ez; transceiver probes carry emitter and detector on one tip.

Electrical route · measured

Echoes from a copper line on silicon.

Test structure: a Cu-based coplanar waveguide on a silicon chip. The THz TDR microprobe carries two photoconductive switches — one launches the pulse into the line, the other detects what comes back.

Photograph of the THz TDR microprobe placed on a copper coplanar waveguide on a silicon chip, annotated with the two photoconductive switches, bias, transimpedance amplifier, pulse propagation and reflection directions.
THz TDR microprobe on a Cu coplanar waveguide (CPW). PC switch 1 launches the pulse in both directions along the line; PC switch 2 samples the returning echoes.
Normalized amplitude of the probe-integrated THz pulse versus time, showing 2.0 ps full width at half maximum and 0.6 ps rise time.
Probe-integrated pulse without DUT: 0.6 ps rise time, 2.0 ps FWHM.
Raw TDR trace over time and distance to fault: a dominant probe-internal transmission peak at zero, followed by small reflections from the first open end between 1 and 2 mm and from the second open end beyond 3.5 mm.
Raw trace. The dominant peak is the probe-internal transmission, followed by echoes from the first and second open end of the line. The distance axis follows from the group velocity.
TDR echo traces for successive probe positions: echo amplitude decreases with propagation distance; extracted attenuation 0.637 dB per millimetre and group velocity 121.2 micrometres per picosecond.
Internal pulse suppressed. Spatially modulating the probe position removes the probe-internal pulse. Extracted on this line: α = 0.64 dB/mm, vg = 121 µm/ps.

Localizing and classifying the defect.

Artificial defects on test structures. The sign of the echo gives the defect type directly; its delay gives the distance to fault (DTF).

Two TDR difference traces at a distance to fault of 1 mm: an open end gives a positive main echo, a full short-cut gives a negative main echo at the same delay; microscope photos show the respective test structures.
Open vs. short. Same distance to fault (1 mm), reversed echo polarity — as transmission-line theory predicts.
Microscope photo of a coplanar line with a semi short-cut and a full short-cut further along, and the corresponding TDR trace showing two separate negative echoes.
Sequenced and gradual defects. A partial (semi) short-cut and a full short-cut further along the line, resolved in one trace.
121 µm/ps
Group velocity

Measured on the Cu-CPW; converts echo delay to distance.

0.64 dB/mm
Line attenuation

Extracted from the echo decay along the line.

1 mm
Demonstrated DTF

Open, short and semi-short located at this distance.

≈ 4.5 mm
Echo range

Echoes detected up to this distance on the lossy on-chip line.

Sensitivity · accuracy · limits

What THz TDR does — and where it stops.

Numbers derived from the measured rise time and the group velocity on the Cu-CPW test line. Absolute sensitivity depends on coupling and device, and has to be determined on your structures.

Resolution & accuracy
  • Two-event resolution ≈ vg·tr/2 ≈ 35–40 µm (tr = 0.6 ps, vg ≈ 120 µm/ps).
  • Localization accuracy is far better than the resolution: the optical delay is set with fs precision (50 fs ≙ 3 µm).
  • In practice limited by the knowledge of vg / εeff: 1 % error in vg ≙ 10 µm at 1 mm distance.
Sensitivity
  • Opens, shorts and partial shorts detected on test lines; type from echo polarity.
  • Discontinuities shorter than vg·tr appear with reduced amplitude.
  • SNR scales with averaging (∝ √N); golden-trace subtraction removes static structure.
Limitations
  • Loss and dispersion broaden the pulse: resolution degrades with distance; practical range is a few mm on lossy on-chip lines.
  • The probe needs a launch point on the line, and the probe-internal pulse must be suppressed.
  • Branched nets and multiple discontinuities overlap; modelling or a golden sample is needed.
  • Lab-proven on test structures — not yet validated on production packages, throughput not yet optimized.
Structural route · measured

Looking through the mold compound.

Near-field imaging in transmission gives fast lateral contrast; reflection adds depth. Both work contactless on assembled parts.

Reflection · assembled module

Depth from time of flight.

A transceiver probe is raster-scanned above a graphics-card module. Different time windows of the reflected transient address different depths — package surface, internal lead and bond structures, die area.

Illustration of a TeraSpike transceiver probe above the main IC of a graphics-card module, indicating near-field reflection measurement.
THz reflection amplitude map of the module, earliest time window.
Gate 1 · earliest
THz reflection amplitude map of the module, second time window, die area strongly reflecting.
Gate 2
THz reflection amplitude map of the module, third time window, lead frames of the memory chips visible.
Gate 3
THz reflection amplitude map of the module, latest time window, internal structure of the main IC package visible.
Gate 4 · latest

One scan, four time windows of the reflected transient. Field of view about 65 × 85 mm². Demonstration sample: a commercial graphics card.

Transmission · IC authenticity

Hidden deviations between identical-looking ICs.

An original and a counterfeit IC, externally indistinguishable. Amplitude and phase maps at 1.2 THz resolve clear deviations in the die area and the lead structure — consistent with the X-ray reference.

Mode
Transmission, scan from the bottom
Analysis
FFT maps at 1.2 THz, amplitude & phase
Pixel pitch
50 µm
Limitation
Integrates over the full package depth — no depth information
Original and counterfeit IC in top and bottom view; THz frequency-amplitude and phase maps at 1.21 THz for both samples with arrows marking deviations in the die area and lead structure; corresponding X-ray images of the lead frames.
Left: the two ICs. Centre: THz amplitude (top) and phase (middle) maps; bottom: X-ray reference. Arrows mark the deviations. X-ray images: F.I.C.S., University of Florida.
Reflection + sparse deconvolution · 3D

A 3D view into the package.

Sparse deconvolution turns the reflection transients into depth-resolved reflector maps: die position, pads and bond wires. The die of the counterfeit sits at a different depth — information not accessible in transmission.

Probe
Transceiver, ~100 µm above the IC
System
Fiber-based TDS at 1550 nm + SHG, LT-GaAs switches
Scan
8 × 15 mm², 100 µm pitch
Acquisition
50 ps window, 50 fs step, 180 averages
For two IC samples: X-ray image, THz bottom view, side view showing the packaging surface, metallic pads and the lower die area at different depths, and an angled 3D point-cloud view with bond wires marked.
X-ray, THz bottom view, side view and angled 3D view for both samples. Black: metallic pads inside the package; purple: lower die area, differing significantly between the samples; red: bond wires from the pads to the die. X-ray images: F.I.C.S., University of Florida.
Proven in the field

From the lab to the fab.

Samsung Electronics First in-line deployment of near-field THz-TDS with photoconductive microprobes in a semiconductor fab: electrical metrology of memory integration processes — tungsten film sheet resistance and nitrogen at the gate-oxide/Si interface. Jun et al., Commun. Eng. 4, 30 (2025). In-line · wafer University of Florida (F.I.C.S.) Florida Institute for Cybersecurity Research uses a TeraCube M2-R scanner for chip-integrity and IC-authenticity testing. Hardware assurance

Wafer-level sheet-resistance mapping of conductive layers →

Further reading

All publications using our technology →

What it takes

The instruments behind it.

Probe, scanner, analysis — configured for transmission, reflection or TDR.

Probes

TeraSpike transceiver & TDR probes

Emitter and detector on one tip → System

TeraCube M2 scanner

Transmission & reflection →
Software

THz data analysis software

Waveforms to maps and depth slices →
Two ways in

Start with your structure.

Whether THz fits your failure-analysis flow depends on the package, the line and the defect. The fastest way to find out is a measurement.

Service

Send us a sample

We measure your device or test vehicle in Aachen, deliver maps, traces and raw waveforms, and tell you what the data does and does not support.

Request a measurement
Feasibility study

Work on it together

For new applications: mutual NDA, your use cases and test structures, a feasibility study at Protemics — and a clear statement of where the method ends.

Talk to an engineer