Skip to content

Vector Magnetic Current Imaging of an 8 nm Process Node Chip and 3D Current Distributions Using the Quantum Diamond Microscope

MetadataDetails
Publication Date2021-10-28
JournalProceedings - International Symposium for Testing and Failure Analysis
AuthorsSean M. Oliver, Dmitro Martynowych, Matthew J. Turner, David A. Hopper, Ronald L. Walsworth
InstitutionsHarvard University, University of Maryland, College Park
Citations9
AnalysisFull AI Review Included

The Quantum Diamond Microscope (QDM) is demonstrated as a novel, non-destructive failure analysis (FA) tool for advanced 3D microelectronics, leveraging Nitrogen-Vacancy (NV) centers in diamond for vector magnetic field imaging.

  • Vector Field Imaging: The QDM simultaneously measures all three magnetic field components (Bx, By, Bz) over a wide field-of-view (4 x 4 mm2) under ambient conditions, overcoming limitations of traditional magnetometers (SQUID, GMR) restricted primarily to Bz and requiring cryogenics/vacuum.
  • High Resolution on Advanced Nodes: The system successfully resolves current paths in an 8 nm process node NVIDIA GA106 GPU, achieving an effective standoff distance of approximately 0.85 ”m and resolving adjacent traces separated by only 8 ”m.
  • 3D Current Path Detection: The QDM’s vector capability enables the detection of vertically oriented current paths (vias) in a custom multi-layer Printed Circuit Board (PCB) analog, which are invisible to Bz-only sensors.
  • Depth Localization Demonstrated: Utilizing the magnetic field profile broadening (Bz component), the relative vertical separation between conducting layers was accurately determined to be ~155 ”m, consistent with PCB design specifications.
  • Layer Isolation Protocol: The linearity of Maxwell’s equations was exploited to subtract magnetic field contributions from known layers (L1 and L3) to isolate and map the current density of an embedded middle layer (L2), demonstrating a path toward 3D current reconstruction.
  • FA Utility: The technique provides a method for in-situ image registration using C4 bump magnetic gradients and shows potential for detecting shorts, leakages, and open faults in complex 3D integrated circuits (ICs) where optical access is blocked.
ParameterValueUnitContext
Process Node (DUT)8nmNVIDIA GA106 GPU
Remaining Silicon Thickness (RST)5”mGPU die thinning for improved resolution
NV Layer Thickness1.7”mDiamond sensor layer
Average Standoff Distance~0.85”mHalf the NV layer thickness from DUT surface
Lateral Spatial Resolution~1”mAchieved resolution for magnetic field mapping
Field-of-View (FOV)4 x 4mm2Simultaneous imaging area
Adjacent Trace Separation8”mMinimum separation resolved on GPU
Trace Width (GPU)11”mJTAG current paths (TRST_N, TDO)
Detected B-Field Magnitude~1”TMagnetic field from GPU current traces
QDM Sensitivity (NV Ensemble)~10pT/√HzTypical sensitivity (compared to SQUID: ~pT/√Hz)
PCB Total Thickness290”mCustom 4-layer 3D analog structure
PCB Layer Separation (L1 to L3)~155”mDetermined via Bz profile broadening
PCB Via Diameter152”mVertical current path dimensions
Excitation Laser Wavelength532nmCW illumination source
Excitation Laser Power1.5WContinuous wave power
MW Power1WDelivered to NVs via loop antenna
NV Purity ([15N])~17ppmNitrogen concentration in diamond layer
  1. QDM Setup and Sensor Placement: A 4x4x0.5 mm3 diamond chip containing a 1.7 ”m thick layer of NV centers is placed directly on top of the Device Under Test (DUT) to minimize the sensor standoff distance (average ~0.85 ”m).
  2. Optically Detected Magnetic Resonance (ODMR): The NV centers are excited using a 532 nm laser (1.5 W CW). Red fluorescence (637-800 nm) is collected via a microscope objective and imaged onto a CMOS camera. Microwave (MW) frequency is swept (1 W power) to find resonance dips in fluorescence intensity, which are proportional to the magnetic field parallel to the NV axes.
  3. Vector Field Measurement: Due to the four possible orientations of the NV axis in the diamond lattice, the ODMR spectrum provides four frequency separations. Fitting these spectra allows for the simultaneous calculation of the Bx, By, and Bz components of the external magnetic field at each pixel.
  4. Sample Preparation (GPU): The NVIDIA GA106 GPU die was thinned to 5 ”m Remaining Silicon Thickness (RST) to further reduce the distance between the NV sensor and the internal current traces.
  5. Background Subtraction and Registration: Large magnetic field gradients caused by magnetic material in the flip chip’s C4 bumps are removed by subtracting the magnetic field image of the unbiased chip (0 V applied) from the image of the biased chip (1 V applied). These C4 bump features are also used for image registration onto provided CAD drawings.
  6. 3D Layer Isolation (PCB): Current is driven independently through specific layers (L1, L2, L3) of the custom 4-layer PCB. To isolate the magnetic field contribution of the middle layer (L2), the magnetic field images of the top (L1) and bottom (L3) layers are subtracted from the magnetic field image of the total structure (L1+L2+L3 active), leveraging the linearity of Maxwell’s equations.
  7. Current Density Mapping: The resultant isolated 2D magnetic field image (e.g., for L2) is inverted using the Fourier Filter formalism to generate a 2D current density map, confirming the location and path of the embedded current trace.

The Quantum Diamond Microscope (QDM) technology is highly relevant for advanced microelectronics manufacturing and failure analysis, particularly in the context of 3D packaging and scaling beyond Moore’s law.

  • 3D Microelectronics Failure Analysis (FA):

    • Detection and localization of shorts, leakages, and open faults in complex 3D structures (e.g., those using Through-Silicon Vias (TSVs) or Cu-Cu connections).
    • Mapping current paths in multi-stack die where traditional optical methods (LIT, OBIRCH) are blocked by backside metallization or dense metal layers.
    • Identifying vertical current paths (vias) that are undetectable by Bz-only magnetometers (SQUID, GMR).
  • Advanced IC Process Nodes:

    • Characterization and FA for chips utilizing backside power delivery (e.g., Intel PowerVia) and beyond 5 nm node CMOS, where active layers are buried deep beneath silicon and metallization.
  • Non-Destructive Testing (NDT):

    • Providing high-resolution, wide field-of-view current mapping under ambient conditions, enabling faster throughput compared to cryogenic, vacuum-dependent, or point-scanning techniques.
  • Metrology and Design Verification:

    • Verifying current distribution and path integrity in complex multi-layer PCBs and ICs during the design and fabrication stages.
    • Providing quantitative z-depth information for current sources, crucial for diagnosing faults in vertically integrated packages.
View Original Abstract

Abstract The adoption of 3D packaging technology necessitates the development of new approaches to failure electronic device analysis. To that end, our team is developing a tool called the quantum diamond microscope (QDM) that leverages an ensemble of nitrogen vacancy (NV) centers in diamond, achieving vector magnetic imaging with a wide field-of-view and high spatial resolution under ambient conditions. Here, we present the QDM measurement of 2D current distributions in an 8-nm flip chip IC and 3D current distributions in a multi-layer PCB. Magnetic field emanations from the C4 bumps in the flip chip dominate the QDM measurements, but these prove to be useful for image registration and can be subtracted to resolve adjacent current traces in the die at the micron scale. Vias in 3D ICs display only Bx and By magnetic fields due to their vertical orientation and are difficult to detect with magnetometers that only measure the Bz component (orthogonal to the IC surface). Using the multi-layer PCB, we show that the QDM’s ability to simultaneously measure Bx, By, and Bz is advantageous for resolving magnetic fields from vias as current passes between layers. We also show how spacing between conducting layers is determined by magnetic field images and how it agrees with the design specifications of the PCB. In our initial efforts to provide further z-depth information for current sources in complex 3D circuits, we show how magnetic field images of individual layers can be subtracted from the magnetic field image of the total structure. This allows for isolation of signal layers and can be used to map embedded current paths via solution of the 2D magnetic inverse. In addition, the paper also discusses the use of neural networks to identify 2D current distributions and its potential for analyzing 3D structures.