1.1.

History of THz Adoption, Improvements, and Applications

1.2.

THz Inspection for Research and Development of New Technologies

THz inspection for semiconductor materials can be generalized into counterfeit detection of ICs via spectroscopy or signal processing, failure analysis (FA) of circuitry and components known as defect localization, thin film measurements using THz spectroscopy, and solar cell characterization for third-generation solar panels.

1.2.1.

2D materials

THz radiation is closely linked to cutting edge research on fabrication of nanostructured semiconductor materials³⁵ from thin film two-dimensional (2D) materials such as graphene to advanced solar cells such as perovskite.^36–43 These THz inspection advances and their physical understandings are critical to the development, characterization, and implementation of new technologies for transistors, p-n junctions, and sensors. However, 2D materials are not yet produced in high volume manufacturing (HVM) and are thus of less interest for the NDT inspection and hardware security industry.

Research and development methods for cutting edge device FA often involve destructive methods, where THz is less advantageous over other inspection methods for its ability to penetrate samples. As advanced 2D materials become viable competitors for gate all around (GAA) and NW-based transistor technologies, advanced THz methods, will augment or replace critical metrology tools for in-line monitoring of the thousands of process steps used in fabrication of billions of connected transistors. The integration of 2D materials into current semiconductor fabrication lines is already being successfully explored at scale and low cost for graphene in 2019,⁴⁴ with the possibilities of graphene technologies for electronics detailed in Fig. 1. The convergence of THz radiation and nanotechnologies, such as graphene, was reviewed by Lawler et al.⁴⁵

Fig. 1

Potential application fields for graphene and related 2D materials where wafer-scale integration is required. Applications addressed mainly by graphene are colored light red and applications addressed by TMDs are colored green. RF, radiofrequency; EMI, electromagnetic interference. Reprinted with permission from Ref. 44 Springer: Nature Materials.

1.3.

Scope

Terahertz (THz) technology is a rapidly advancing field, and there have been many reviews covering industrial, medical, and research uses.^1,50,51 This review will present and highlight traditionally used semiconductor inspection methods and detail how THz methods compare in terms of advantages and disadvantages for the near future. There are many well-cited and comprehensive reviews on individual semiconductor devices. THz optoelectronics covered many useful THz topics in 2005, but specifically THz radiation from semiconductor surfaces were presented in detail by Gu and Tani.⁵² Tonouchi⁵³ highlighted THz applications semiconductor and other industrial applications for security and medicine.

In 2011, carrier dynamics in semiconductors studied with time-resolved THz spectroscopy was presented by Ulbricht et al. highlighting applied inspection methods for determining spectral signatures of different semiconductor quasiparticles, phonon resonances, and intraband transitions in low-dimensional systems. THz was displayed as an optimal contact-free and noninvasive method suited to probe the conductivity of, in particular, nanostructured materials that are difficult or impossible to access with other inspection methods.⁵⁴

In 2017, the THz roadmap was established by an invited a group of international experts detailing 18 sections that cover most of the key areas of THz science and technology.⁵⁵ Important sections relevant to semiconductor inspection are summarized as:

Schmuttenmaer presented THz spectroscopy of semiconductors and semiconductor nanostructures, covering single particle studies, broad bandwidth limitations, spatial resolution, nonlinear optics, and the challenges to overcome.

Zeitler et al. covered NDT and molecular spectroscopy, detailing the current challenges for THz-TDS inspection due to penetration depth and hardware-limited imaging systems.

Cumming et al. detailed the components for THz imaging, and the hardware such as CW, pulsed, and focal plane arrays and the limitations for scaling systems into the real world.

Castro-Camus reviewed photoconductive devices for THz time-domain spectroscopy; the history of how the bandwidth, power, and reliability of these devices improved with the use of semiconductor materials and contact structures and what challenges lie ahead such as fabrication cost.

PCA-based devices are fabricated using semiconductor device processes such as ion implantation for doping, which can greatly affect the PCA THz signal and spectral quality for spectroscopy.⁵⁶ Burford and El-Shenawee⁵⁷ provided a collection of literature on the current state and remaining challenges of THz PCA technology as of 2017.

A perspective was shared by Mittleman⁵⁸ detailing the access nonlinear optical effects in the THz range and resultant methods to probe nanoscale phenomena. This perspective translated into a review of THz imaging techniques in 2018, with specific semiconductor applications.⁵⁹ THz imaging and sensing was reviewed in 2019, specifically hardware based on silicon technologies.⁶⁰

In 2019, advances in THz solid-state physics and devices offered an overview of the most active research areas currently under investigation in the broad field of THz solid-state devices, further unveiling the properties of these materials on an ultrafast timescale.⁶¹ In relation to semiconductor inspection, the challenges involved in achieving real world applications are presented, for example, THz NDT inspection of oxidized silicon demonstrating that the potential of a semiconductor covered by an insulator which can be charged by the corona charging setup, used to tune the surface potential of the semiconductor without electrical contacts, can be evaluated successfully by assessing the THz emission.⁶²

A summary of applied near-field techniques for semiconductor materials is presented at the introduction of Sec. 6. Details covering nondestructive technology NDT reviews for THz inspection is covered in Sec. 3.

Bibliometric analysis of the THz field by Li et al. in 2020⁶³ is critical to demonstrate the connections between all fields of THz research. The connections between THz hardware and the applied research can be demonstrated visually and graphically. This is especially helpful and important for understanding how semiconductor fabrication techniques relate to advances in THz hardware devices. A bibliographic review was provided by Lewis⁶⁴ at the introduction for his review on THz source hardware, graphically demonstrating that the number of THz research papers has reached exponential growth. Liu⁶⁵ reviewed over 10,000 papers and 60 patents in 2012, covering THz technology.

Section 1 presents the history for THz NDT electronics inspection and a summary of the existing reviews, perspectives, and roadmaps on semiconductor inspection methods. In Sec. 2, this review will introduce the semiconductor supply chain and the most pressing hardware security vulnerabilities at all device scales: transistor, IC, and printed circuit board. Section 3 will summarize the hardware and terminology for THz NDT inspection in the context of applied electronics inspection. Section 4 will detail currently applied THz inspection methods that are relevant to the FA and hardware verification community at the transistor, IC, and PCB scales.

2.1.

Semiconductor Supply Chain Vulnerabilities

End-user electronics such as smartphones, personal computers, and industrial equipment operate upon the on/off switching of many transistors combined to perform operations. These logical operations of transistors can be designed and fabricated by the front end of line (FEOL) processes shown in Fig. 2(c). These transistors and their back end of line (BEOL) connections can be incorporated into an application-specific integrated circuit (ASIC) for use in an end-user device shown in Fig. 2(b). After many years of innovation, there are many general ASIC designs that have widespread uses and can be bought off the shelf (COTS), colloquially known as ICs. These IC “die” can be mass manufactured upon a semiconductive substrate, often silicon wafer, through many process steps such as deposition, lithography, and etching. This fabrication step is demonstrated as stages 1 and 2 in Fig. 3. After many production steps these IC dies can be cut, stage 3 Fig. 3, from the wafer, packaged, tested, stage 4 Fig. 3, and connected to the outside world. This packaging step is where the epoxy molding, sometimes ceramic for RF applications, encapsulates and protects the die. External connections such as pins, leads, or ball grid arrays (BGA) enable electrical connectivity from the miniature die to other macrocomponents and devices. Stage 5, Fig. 3 represents a vulnerability for modification in the supply chain as devices are sent to inventory, until needed for final integration. Modern devices are often comprised of a series of ICs integrated at stage 6 Fig. 3; with various functions from memory, communication, and processing, all connected together by the copper circuitry and substrate of a PCB shown in Fig. 2(a) ready for distribution, stage 7 Fig. 3.

Fig. 2

(a) Assembled printed circuit board (PCB) composed of multiple passive and active components such as (b) integrated circuits (ICs), seen in cross-section depicting the die connection to the PCB substrate via the C4 solder joint. In (c) Front end of line (FEOL) & Back-end of line (BEOL), a cross section depicts the FEOL transistors connectivity through the metal layer connections to the C4 solder bumps. Figures adapted from Ref. 66.

Fig. 3

Global semiconductor market as of 2016, published in 2018 by SIA adapted from semiconductors.org. Source: UN Comtrade and Taiwan Customs Administration, Ministry of Finance; Year of 2014.

Defects, inclusions, or malicious modifications from the transistor to the assembled device can present a reliability and functionality issue for end-users. Designers implement “design for manufacturing” criteria enabling complex device designs from transistors, ASICs, and multilayer PCBs to be tested effectively during and after production, while mitigating intellectual property (IP) vulnerabilities due to outsourcing. Manufacturers take many steps during production to mitigate defects and prevent faulty devices from reaching market. Test-houses perform inspection to ensure all devices exhibit a specified level of performance and functionality to ensure a device will operate over its entire life cycle.

2.2.

Hardware Threats

2.3.

Hardware Verification Techniques and Countermeasures for Hardware Trojans

The primary goal for semiconductor FA techniques is to identify the root cause of a production error or unexpected inclusions that can alter the expected electrical and reliability performance affecting the overall yield or life cycle operation. More stringent automotive standards from SAE and ISO J3061 and 21434 address cybersecurity for the transportation industry for example. These standards increase the requirements for companywide cybersecurity protocols along with increasing the verification effort on critical components. Across end-user electronics industries from the internet of things to industrial or infrastructure electrical equipment, there is an increase in the verification standards, causing issues for fabricators facing production slowdowns.

The characterization of a defect or error is best achieved through NDT methods as the destructive process does not introduce added artifacts or remove defect information. For this reason, typical FA workflows start with NDT methods, and then once the maximum information is collected, more destructive methods are employed such as polishing or sectioning. X-ray imaging and C-mode scanning acoustic microscopy (C-SAM) are examples of currently effective volumetric methods for IC and PCB NDT FA. Destructive methods are based on isolation of the defect location and then methodically remove enough material via FIB, polishing, or etching, to adequately characterize the electrical features of interest via nanoprobing, optical, or E-beam techniques.

The identification of defective connections and proper operation is a focus for manufacturers before an IC die is validated and encapsulated by an epoxy molding. Electronics assembly houses for PCBs focus on ensuring their IC components are legitimate before integration through optical verification of surface text, electrical testing of external pins, or volumetric methods such as x-ray, infrared thermography, surface acoustic waves, scanning acoustic microscopy, or magnetic current imaging. The nanoscale features and layers of ICs and PCBs are often below the spatial resolution of commonly used NDT methods to resolve structural information. This has resulted in the need for destructive cross sectioning to achieve the necessary resolution, but this process introduces damage to the sample possibly limiting characterization. IC die and advanced package feature sizes and their defects are no longer able to be characterized with SEM alone and must be analyzed via complex FIB/TEM FA workflows. Complex multilayer PCBs are not able to be characterized with x-ray due to metallic noise and streaking during imaging due to tin solder, copper wiring, and aluminum surface-mounted components. While destructive approaches are able to verify an entire device with confidence, these are not scalable to all devices, NDT are effective but increased device density has limited spatial resolution of volumetric methods. Design for security implements additional costs for production, while run time monitoring is unable to protect against all threats. A depiction of a taxonomy on countermeasures to mitigate hardware defects or modifications is shown in Fig. 4.

Fig. 4

Overview of different protection approaches against hardware Trojan attacks.⁷³ Copyright © 2014 IEEE. Reprinted with permission from Proceedings of the IEEE.

2.3.1.

Presilicon

In a perfect verification workflow, an intrusion should be detected before it is able to manifest into failure of the device or a hardware Trojan attack. During presilicon testing, it is possible to detect an intrusion in the circuit design during the design verification process through testing the functionality of a design before sending for production using simulation software such as EDA tools or through programmable hardware such as Field Programmable Gate Arrays. This functional verification then goes onto physical design and verification where the foundry determines if the functional design can be feasible fabricated via a prototype recipe that passes device performance specifications. Although presilicon detection might appear to prevent end-user data leakage or system maloperation, a hardware Trojan HT may be configured to avoid presilicon testing altogether. For example, an HT may be inserted in an IP core provided as an RTL model where it is only designed to activate after it has been shipped to the end-user via a payload or timed Trojan.

Fig. 5

List of current imaging techniques used for semiconductor components. (a) Imaging speed: the time cost of taking one image. (b) Magnification. (c) Image dimension. (d) Cost is estimated based on the online available commercial products. (*) Price provided by Google Shopping. Table adapted from Ref. 75.

2.4.

Postsilicon

These methods include manufacturing test, silicon validation, and system test and validation, where issues during production are detected. Postsilicon methods are used to validate the final assembly of a device. This is not performed with one test method but occurs as part of a manufacturing testing workflow to prevent cascading of errors during process steps. These postsilicon methods often include in-line (part of the assembly line) inspection tools that can nondestructively analyze the success of the previous production step and quickly detect any defects or errors without limiting the speed of the “line” (Fig. 5). Due to the fast-paced production speed of a high-volume manufacturing, the in-line tests must be capable of high confidence at high speeds. In-line testing prevents process errors and characterizes device quality and performance. Off-line methods, outside of the production workflow, are required to find the root cause and mitigate an error found by in-line testing or after production. These can include destructive methods to isolate a defective region through cross sectioning to achieve better access for a metrology method such as SEM, TEM, or electrical probing. Final postsilicon testing might involve the random selection of a statistically significant portion of a batch for destructive testing to ensure the quality of the rest of the untested batch. For this reason, postsilicon test methods that incorporate testing of each device enable a higher level of confidence in device functionality. While the design of an IC device is still mainly performed by entities in North America, the fabrication often occurs in the untrusted environment of the Asia-Pacific region. Presilicon can be used to protect malicious design but not prevent production issues, for this reason postsilicon testing has become a requirement due to the numerous vulnerabilities during the production phase. Tables of common semiconductor test methods at the PCB in Fig. 27, IC NDT methods in Fig. 25, and transistor scale inspection methods for ICs in Fig. 26 are provided in the Appendix.

3.1.

Pulse Versus Continuous Wave THz Generation

There are various techniques for the generation of THZ radiation, but generation is dominated by photoconduction and optical rectification methods. In the photoconductive approach, high-speed photoconductors are used as sources for antennas, such as large-aperture photoconducting antennas. The optical rectification approach uses electro-optic crystals as a rectification medium and depending on the optical fluence the rectification is a second-order (difference frequency generation) or a higher-order nonlinear optical process.⁹⁴ These two different methods of generation result in differences in the bandwidth, speed, and resolution. The various THz sources as of 2014 are detailed by RA Lewis’ along with a bibliometric perspective of published THz research over recent years.⁶⁴ Although there are many hardware methods for generation, very few can generate the peak intensities required for nonlinear optics experiments and analysis.⁵⁸ Pulsed systems offer TDS capabilities, where the time delay can be used to determine phase and amplitude from a sample. This can enable determination of thickness of layers, absorption properties, dielectric constants, and refractive index simultaneously.

Pulsed systems offer advantages in wide bandwidth volumetric inspection over CW systems which themselves benefit themselves from lower cost hardware and increased spectral signal quality, with more variability in the modes of generation. The generation of THz plays a large role in what is detectable using reflection or transmission mode geometries. The detection of THz radiation is as sophisticated as the generation, and this is a function of the requirement to detect THz energies on the orders of millielectronvolts, THz wavelengths on the millimeter/submillimeter scale and THz wavenumbers amounting to tens or hundreds of inverse centimeters. In 2019, the current state-of-the-art of detector technology was reviewed, while relating detectors to their informed applications.⁹⁵ An optimal THz detector is able to detect individual THz photons; however, this is difficult in practice due to inherent noise in the atmosphere and electromagnetic spectrum. A detector is often more responsive to specific frequencies, and this can be seen among THz systems as the SNR decreases drastically above 4 THz.

Pulsed time-domain measurement and CW THz measurement have different strengths and limitations of the modalities for NDT applications. It is decided that CW imaging allows for a more compact and simple system, while pulsed measurements yield a broader range of information.⁹⁶

3.2.

Reflective and Transmission Measurements

The low power energy of THz sources does not allow for a powerful signal reflection or transmission through objects, even those comprised of THz transparent materials. THz reflection and transmission imaging of insulated copper cables to determine corrosion behind packaging for infrastructure component analysis and detection of security threads in banknotes.⁹⁸ This increases the dwelling time for THz imaging techniques. The absorption of THz radiation or the diffraction of the THz beam can limit the resolution to the size of the wavelength of the beam. Similar to how transmission-based metrology such as TEM offers increased spatial resolution as opposed to scattering methods such as SEM, THz transmission imaging offers increased resolution if the adequate THz radiation can pass through the object as opposed to reflective modes (Fig. 7). Transmission or reflectance collection is possible in the near field or the far field, while near-field imaging is limited to thin samples.

3.3.

Near-Field and Far-Field THz Techniques

Gurtler et al. presented a comparison of near-field and far-field techniques in 2000, comparing that the ultrashort THz pulses emitted from a biased, large-aperture THz antenna between an unfocused THz beam and a focused THz beam, finding that the THz beam essentially behaves like a fundamental Gaussian beam with a beam waist given by the spot size of the emitting laser pulse at the THz emitter. This remained true for other types of large-aperture emitters, since the propagation characteristics are independent of the microscopic processes generating the THz beam.⁹⁹

TDS sampling methods are based on collection of the THz field with a femtosecond laser pulse that the propagation characteristics are understood are reproducible. A convolution technique is used to pair the source’s pulse to the detected response. There are many techniques to convolve these signals but is based on the principle of characterizing the THz frequency received as opposed to just the amplitude of the signal received. The pulse contains multiple bandwidths of THz, with the strongest signals being in the 0.5- to 2.0-THz ranges but extending up to 10 or 50 THz is some cases with adequate signal quality. The difference between TDS and CW systems is that frequency TDS-based methods are able to characterize the entire THz bandwidth with femtosecond second repetition compared to the intensity-based methods such as CW which can only characterize phase or intensity. TDS systems combine wide bandwidth frequency information with time-domain information to analyze samples in three dimensions by correlating depth information to the time-domain pulse’s reflections which indicate interfaces of layers. CW systems are more effective for their single frequency dynamic range capabilities to analyze a specific THz wavelength, even capable of higher spectral resolution than TDs systems.

The benefit of near-field imaging is to collect the high spatial frequency information typically lost due to decay of THz signals as they propagate from the surface of a sample depicted in Fig. 8. THz research has been focused on increasing the spatial and spectral resolution of THz NDT methods past the limits of the diffraction. Reconstruction methods based on TDS measurements have been used to develop algorithmic-based image restoration techniques in the far-field region, while scattering-based methods have been used to develop image restoration for the near-field region.

Fig. 8

In near-field (Fresnel) wave fronts are considered close to spherical whereas in the far-field the wave fronts are essentially planar (Fraunhofer). Fresnel diffraction or near-field diffraction is a process of diffraction that occurs when a wave passes through an aperture and diffracts in the near field, and if the distance is increased, outgoing diffracted waves become planar and Fraunhofer diffraction occurs.

Pulsed near-field methods are used for microscopy or nanoscopy have been used for subwavelength imaging since 2008.¹⁰⁰ These near-field imaging methods require raster scanning which is time consuming and are based on a probe that is brought in close proximity to the surface of a sample. The probe and sample interact in a way enabling detection of radiation by the probe if equipped with an antenna or scattered to a coupled detector. The resolution limitation of near-field probe techniques is by the fabrication quality and feature size of the probe tip, which is smaller than the wavelength of THz radiation. Current methods offer real-time imaging in the near field using single pixel detectors, which do not require raster scanning eliminating the need for linear $xy$ stages.¹⁰¹

3.4.

THz Imaging

THz ability to penetrate plastics and ceramics makes it an excellent choice for imaging of packaging materials, while offering 3D spectral information about an object through THz pulses in the time domain. With the combination of more capable hardware and mathematical phase retrieval methods, THz could be used to rapidly image and characterize complex objects such as ICs consisting of many different material compositions. CW THz imaging involves raster scanning of a beam or the rectification of signals onto a detector array with adequate incident radiation. For rectification, a single beam can be used to develop the full amount of radiation power needed for signal quality and benefits from a high dynamic range, but the raster scan must collect information in a slow speed point by point. As opposed to a TDS inspection tool, CW system can outperform systems based on time-domain photoconductive sampling in frequency resolution, spectral brightness, system size, and cost. Figure 16 shows a THz image collected by a TDS system, but depicts an isolated 1.996-THz frequency which is mostly transparent to the IC packaging. Panchromatic absorption is where the absorption coefficient is collected over entire frequency bandwidth of pulse and is then calculated through the change in the amplitude of the reflected or transmitted pulse. Monochromatic-absorption, in comparison is where the coefficient is measured at a defined frequency or a limited range of frequency. Jepsen details a “how to” tutorial on derivation of phase from TDS measurements, THz time-domain spectroscopy (THz-TDS) is in many ways a well-established, proven, and versatile spectroscopic technique that is frequently and routinely used in many laboratories. The basis of high-quality optical data on materials using THz-TDS is the correct extraction of the complex-valued dielectric properties (index of refraction, permittivity, or conductivity) from the recorded amplitude and phase of the involved THz signals.¹⁰² Guerbouhka et al. reviewed recent advances in THz imaging with an emphasis on the modalities that could enable real-time high-resolution imaging. This covers several key imaging modalities developed over the years such as THz transmission, reflection, and conductivity imaging; In addition to THz pulsed imaging, THz computed tomography (CT), and THz near-field imaging techniques. All of these imaging modalities are part of the enabling technologies for real-time THz imaging within the time-domain spectroscopy paradigm which will require fast optical delay lines, photoconductive antenna arrays, and electro-optic sampling with cameras.¹⁰³ From the THZ camera technology to the real-time methods that will enable live THz imaging with single-pixel detectors, mechanical beam-steering, compressive sensing, spectral encoding, and fast Fourier optics are also detailed as of 2018.

4.1.

PCB and THz Radiation Interaction

PCBs are currently used as a platform for THz on chip development,^110–112 enabling subwavelength sensing to allow fully integrated spectroscopic chips operating across 0.04 to 0.99 THz without the need for any external THz source, drastically lowering costs for THz inspection systems. For this reason, there is extensive research into the THz radiation penetration depth and effect upon FR4,¹¹³ the epoxy woven fiber material which comprises the substrate of a PCB. PCBs can often be used as motherboards for desktops or logic boards for smartphones and the understandings of how THz signals propagate through or near a PCB are critical. The characterization of the effects of 300 GHz (0.3 THz) radiation on PCBs indicates that the presence of the ground plane or parallel-plate structures in the channel introduces multipaths that, if constructively superimposed, may create path loss lower than the free-space propagation path loss.¹¹⁴ In other words, the internal copper ground plane and other parallel structures can act as artificial antenna enhancing THz radiation to propagate. Therefore, the design of a PCB can constructively or destructively interfere with THz-based characterization methods between ground planes or parallel structures. Copper traces and metallic solder are highly attenuated in THz imaging techniques allowing for proper characterization of circuitry shorts, spurs, and opens. Due to the highly attenuating nature of copper for THz radiation, during compressed sensing experiments PCBs copper patterns are used, where the “copper” pixel corresponds to pixel value zero on the random pattern, while a pixel without copper corresponds to the value one, since the PCB material (FR4) is fairly transparent to the THz beam.¹¹⁵ PCB substrate material strongly absorbs THz radiation at frequencies above 1 THz¹¹⁶ which can be seen in Fig. 9 and glass fiber epoxy might exhibit anisotropic effects on THz signals.¹¹⁸ Due to this absorption, it is difficult to reconstruct THz images with high quality while keeping the spectral information, and therefore absorption above 1 THz currently limits PCB imaging to near-field methods, or large frame averaging volumetric methods, causing large time requirements for data collection. For example, T-ray imaging has been applied to PCB failure detection in a passive manner, specifically on the hardest to detect of defects where test program sets are those that lead to ReTest Okay situations. ReTest Okay situation refers to a failed test, where the defect was unique to a sample and not across a lot or batch. These defective cases where the cause is not repeatable cannot be resolved with electronic testing alone and yet certainly have an underlying physical defect as a cause.¹⁸ THz-TDS has been used to detect hidden inclusions with various shapes and materials in glass fiber-reinforced polymer (GFRP) composite solid panels which are similar in structure to a PCB substrate. In 2019, traditional transmission and reflection THz-TDS imaging are performed for detecting GFRP inclusions while comparing spectral difference between defect area and nondefect area and dynamically adjusting imaging parameters to the defect characteristics, which greatly improves defect detection.¹¹⁹ The THz-TDS imaging results are compared with those employing x-ray CT and C-SAM.

Fig. 9

Transmission of the imaging area within the metallic ring (blue diamonds) and of a standard PCB substrate (red circles).¹¹⁷

Signal processing of THz radiation is often performed through modeling of a THz beam. This is achieved by incorporating an understanding of the beam’s interaction with a sample. To accurately determine material properties of a PCB, this will require an understanding of how a THz signal propagates through a PCB structure. It has been demonstrated that the scattering model for the reflection of broadband THz signals from the surface of PCB and its straightforward structure and application to any broadband ray-tracing algorithm is simple.¹²⁰ This simulated scattering model can replicate results with a signal level and time-domain shape very close to that of actual THz measurements and their response from a PCB. This is important as the modeling of THz sample interactions enable higher spatial resolution characterization.

Further analysis of plastics materials commonly found in IC and PCB packaging determined two takeaways: first, transmission is a nonnegligible effect when it comes to plastic materials. This needs to be considered, e.g., for considerations on electromagnetic compatibility. Second, reflection from plastic materials is heavily influenced by the multilayer structure of a material. Thus, significant pulse-broadening may occur even at small bandwidths if this behavior is not taken into consideration.¹²¹ The understanding of THz effect on PCB materials such as FR4 and copper can be used for modeling and developing appropriate metrology methods for PCB substrate, passive components, and circuitry.

Figure 10 shows the nondestructive detection imaging on a Xillinx PCB covered by a plastic case with a scanning step length set to 1.0 mm. The original THz image is shown in Fig. 10 with a size of $101\times 38$. It can be seen from Fig. 10 that the red frame marked area looks like several connected electronic components, but the truth is there are four separated electronic components, as shown in Fig. 10. After doing super-resolution reconstruction with the proposed method, the four separated electronic components can be clearly distinguished, as shown in Fig. 10. The effects of various algorithms can cause the electronic components to look larger than the truth due to the blurred edges.³³

Fig. 10

Experimental results on real THz image of circuit board with scale factor 3. The red frame marked area in the lower left corner of each image is the corresponding zoom-in image. (a) Original image, (b) bicubic interpolation, (c) Lucy–Richardson and bicubic interpolation, (d) VDSR, and (e) proposed method. NDT imaging on a circuit board covered by a plastic case shown in (f) circuit board and (g) plastic case. Adapted with permission from Ref. 33 © The Optical Society.

4.2.

THz Computed Tomography

A review of THz techniques was presented by Guillet et al. in 2014,¹²⁴ covering the principles of tomography for THz CT, tomosynthesis, synthetic aperture radar, and time-of-flight THz tomography along with advantages, drawbacks, and limitations for imaging the internal structure of an object.

THz CT first adopted in 2002,¹²⁵ building upon Mittlemans’s standard T-ray tomography has been widely adopted for many applications¹²⁴ but is primarily limited by a sample’s thickness as the low power of THz sources does not enable large transmission percentages for THz radiation. X-ray microfocus CT is currently used to perform volumetric inspection of PCBs in an off-line FA workflow¹⁰⁸ X-ray micro-CT offers superior volumetric capabilities when compared to THz CT. However, THz CT has superior spectral characterization compared to x-ray spectral capabilities, where flat panel detectors are limited by the conversion of x-ray radiation into visible light. Moreover, THz spectral CT can provide colored pictures in THz range as compared to monochrome x-ray CT, associated to the possibility to tentatively analyze the chemical composition of the sample based on THz spectral fingerprints.¹²⁶

Time domain-THz CT 3D imaging has been proposed by NASA for commercial application in PCB delamination detection and can be advantageous when compared to traditional tomographic imaging which has substantial difficulty in determining the layer index of refraction and absorption properties without ambiguity.^52,127 CT methods enable generation of cross-sectional slices enabling accurate layer by layer inspection, and THz CT can enable layer isolation of dielectric properties and delamination. THz CT can incorporate knowledge about an object to reconstruct deviations from expectation based on a modeling. A model as such has been proposed based on a Monte Carlo extension of usual ray tracing methods, which aims to account for the Gaussian intensity profile of THz beam and considers refraction and reflection losses when propagating through an object.¹²⁸

Traditional x-ray inspection of multilayer PCB is limited by large noise due to metallic streaking, and information from multiple 3D layers being contained in a single 2D projection. Surface mounted components can block internal features such as traces and vias among other design information. THz can isolate layers through the time delay of the signal, while x-ray images are a combined sum of an object’s effective density in the direction of the beam. Multiple objects positioned along the plane of the x-ray beam will be combined at the detector.

4.3.

Optical versus THz Imaging

PCBs rely on vias to carry signals in multilayer boards, which can be a challenge to design and manufacture causing defects or reliability issues after thermal stresses.¹²⁹ Vias deliver a route for electrical connections and thermal energy to move between layers. Vias are created by drilling or laser boring through a PCB structure in completion or to a certain layer. Filling a via with conductive material provides for connectivity through plating. This process is very sensitive to chemicals or other substances within the holes, which can eventually cause defects and reliability issues. It is important to characterize these vias for their drill quality along with electrical properties, as there are many filling, plating, and etching steps during PCB production that can lead to a stack up of errors.

Optical imaging is often used to characterize process steps for PCB production, often known as AOI. Zhao et al¹³⁰ compared optical and THz methods of subwavelength characterization for PCB through holes. In Fig. 11, an optical microscope image of multiple through-holes on a PCB plate were collected along with femtosecond pulse imaging using THz. Each hole is about $600\mu \mathrm{m}$ in size. Arrows A, B, and C point at three hole pitches with characterized widths of about $60$, 80, and $75\mu \mathrm{m}$, respectively. (b) THz image of multiple through-holes on PCB plate. (c) Spatial profiles of a series of representative holes retrieved from (a) and (b) along the green dotted line, as red line and blue solid squares, respectively.

Fig. 11

Optical microscope image of multiple through-holes on a PCB plate were collected along with femtosecond pulse imaging using THz. Each hole is about $600\mu \mathrm{m}$ in size. Arrows A, B, and C point at three hole pitches with characterized widths of about 60, 80, and $75\mu \mathrm{m}$, respectively. (b) THz image of multiple through-holes on PCB plate. (c) Spatial profiles of a series of representative holes retrieved from (a) and (b) along the green dotted line, as red line and blue solid squares, respectively.¹³⁰

4.4.

Near-Field THz Imaging for PCB Passive Components

Near-field imaging of a PCBs microwave emission was used to characterize transmission line discontinuities in 2001. Due to the small feature sizes of PCB structures such as traces and vias, the ability to measure fields with a high spatial resolution relative to feature size and in the close vicinity of the board would be required. In practice, digital PCBs have most of their traces laid out in two orthogonal directions and only very short oblique segments. Modeling of how radiation emanates from a PCB is critical to near-field imaging techniques. In Fig. 12, THz near-field imaging using CW source was performed on a PCB, with $18\text{-}\mu \mathrm{m}$-thick copper circuitry and, in some areas, electrically insulating paint. Figure 10 shows the results of THz imaging of two board regions, where (a) and (c) show visual images, and (b) and (d) show THz images corresponding to $\nu =0.5\mathrm{THz}$ ($\lambda =600\mu \mathrm{m}$) and electromagnetic wave polarized along the $x$ axis. The electrical circuit includes elements of subwavelength scale, which are clearly resolved in the THz images even in the case of being covered with electrically insulating paint.¹³¹ PCB imaging through transmission imaging is difficult due to the large absorption of high THz frequencies in the far-field passing through the PCB substrate. Near-field imaging offers collection of THz signals near the surface, enabling a higher spatial resolution of characterization than is possible with far-field imaging techniques. IC packaging has been imaged with far-field methods and in transmission geometries, but the thickness and structure of an IC are more transparent than a PCB structure.

Fig. 12

THz imaging of two board regions, where (a) and (c) represent visual images, and (b) and (d) show THz images corresponding to $\nu =0.5\mathrm{THz}$ ($\lambda =600\mu \mathrm{m}$) and electromagnetic wave polarized along the x-axis. The electrical circuit includes elements of subwavelength scale, which are clearly resolved in the THz images even in the case of being covered with electrically insulating paint.¹³¹

PCB inspection often is primarily performed by manufacturers who have the design information about the circuitry patterns. This prior knowledge about the circuit design is beneficial and critical for THz near-field TDS imaging which offers a potential for the detection of defects in circuitry wiring of a few microns in size through the incorporation of PCB design information with THz signal processing. Using this technique, it is shown in Fig. 13 that the imaging of PCB’s traces obscured by a $115\text{-}\mu \mathrm{m}$-thick silicon wafer with $100\mu \mathrm{m}$ ($\lambda /4$) resolution is possible due to the THz sensitivity to local conductivity.¹³² The limitation preventing a further increase in the resolution lied within the thickness of the silicon-based modulator, which is in direct contact with the object to perform near-field imaging. Moreover, two different sets of patterns, Hadamard and random, were applied, showing the superiority of Hadamard masks to obtain a high spatial resolution.¹³³

Fig. 13

Left: Image polarized effects. (a) Images ($64\times 64$) of circuit board in Fig. 12 with vertical polarization. Pixels are $40\mu \mathrm{m}$. We see that the contrast of each of the individual wires in the circuit depends on the THz polarization, with the highest contrast seen for polarization parallel to the wires. (b)–(e) Images ($64\times 64$) of the square regions in (a). Polarization is shown by the green arrow on the top left corner of each picture. Pixel size is $20\mu \mathrm{m}$, and images have been denoised using the algorithm described in Sec. 4.4. We see that the very subwavelength wiring breaks [marked by circles in (b) and (e)] give rise to transmissive regions in the THz image when the THz polarization is parallel to the wire. In (e), the diagonally orientated wire (indicated by the white arrow) also shows low contrast. Every image has been obtained via a full set of Hadamard masks. (F) and (G) Line plots through the $8\mu \mathrm{m}$ gaps in (e) and (e) with amplitude and space on the vertical and horizontal axis, respectively. The spatial coordinates of the plots are indicated by green rectangles in (b) and (e). Right: Hadamard versus random versus raster imaging (a) circuit board design, where black indicates conducting, metallic regions. The individual wires are $50\mu \mathrm{m}$ in width, and $8\mu \mathrm{m}$ breaks have been introduced at points marked by the letters A and B. (B) Image acquired using raster scanning of a single opaque pixel. (C), (D) Comparison of the same image acquired with a full set of masks derived from random and Hadamard matrices, respectively. (E), (F) Compressed images obtained via random masks, where the number of measurements is (E) 25% and (F) 75% of the total number of pixels (we use a total variation minimization image recovery algorithm). In all images, the THz electric field is polarized horizontally, and the number of pixels is $64\times 64$ with $40\mu \mathrm{m}$ pixels. The signal acquisition time for a single measurement is 500 ms. Because of the considerably larger noise in the measurement, we have scaled the image in (b) by 0.25 to use the same color scale as in (C) and (D). Figure and caption recreated from Ref. 132.

4.5.

PCB Feature Sizes Are Trending Toward IC Scaled Features

NDT methods such as x-ray CT have been a key developer for the PCB industry with the ability to detect faults without time consuming destructive techniques. Various electronics applications from the IoT, 5G communications, GPU developments, and artificial intelligence (AI) require superior I/O density and 3D scaling at the device, package, and system levels. This density increase also drives the need for NDT imaging methods with high spatial resolution and if possible, in 3D. As the scale of features at the PCB trend smaller, their footprint reaches that of the features size for an IC.

X-ray inspection systems have been used for microscopy systems building upon developments from synchrotron technology.¹³⁴ In the past 20 years, major developments in optics, stages, and reconstruction enabled x-ray NDT to become an effective method for IC and PCB FA workflows, submicron and nanoscale XRM enable successful FA outcomes from PCB to die BEOL levels, even in complex samples.¹³⁵ X-ray for critical dimension (CD) is used for analysis at the nanometer scale with proper sample preparation. There is a need to nondestructively acquire 3D images at high resolution from the PCB to IC and transistor scales. THz hardware offers the ability to image in the near-field and far-field to acquire information with nanometer resolution or macroscale. Although x-ray offers superior volumetric analysis, THz can resolve chemical and electrical information simultaneously without the use of ionizing radiation.

PCBs enable ICs to communicate with the outside world; however, the advanced 3D integration has begun to look into replacing the PCB as the electrical carrier. There is significant benefits to integrating an IC direction onto a silicon interposer, because a thermal mismatch occurs between different materials (PCB and IC) because of their different coefficient of thermal expansion CTE. This causes mechanical stress within the object or component or device often at the connections such as BGAs, and then the thermal warpage and the resultant deformation causes failure between the IC package and the PCB. As the feature size of PCBs approach those of ICs, it will become more likely to skip the PCB altogether to avoid CTE issues during bonding.

Based on current applied THz methods to PCB materials, there is a potential for THz inspection to replace drill hole verification, detection of broken circuitry near the surface or at the surface. THz is also able to penetrate plastic packaging to see inside of PCB housings. These NDT methods of PCB verification can be used for reliability and quality assurance of PCB components during and after production.

5.1.

IC Design

IC complexity can be separated by its interconnect type and feature size. Some examples are flip-chip, wirebond, wafer-level packaging (WLP), and through-silicon vias (TSVs) which comprise many of the connections from the macroscale flip-chip IC to PCB connections down to the nanoscale transistor and die connections. A modification or defect at any of these levels can cause device mis operation, data leakage, or premature failure of an IC during its life cycle. This failed IC can cause compromise or enable malfunction of interconnected systems such as the assembled PCB device due to the prematurely failed IC. IC packaging is expected to provide a physical housing and enable reliable signal interconnections from the inside to the outside or vice versa. THz ICs have been developed for communications and sensing drastically lowering costs; however, there are several challenges associated with increased device frequencies and resulting signal loss, dimensional constraints, and microfabrication methods. An overview of interconnections and packaging technologies dealing with these packaging and interconnection issues for THz electronics such as those used for communications or sensing was provided by Song in 2017.¹³⁶ There are many packaging methods such as commercial ceramic technologies, micromachining, and 3D printing technologies discussed as well.

IC integration scales from small- to very large-scale integration, referring to the number of components per chip. As the number of connections increased, and approaches VVLSI, with more than 1 million components per chip, this has required higher spatial resolution and faster techniques to meet the high-volume requirements of the electronics industry. THz IC characterization can be segmented into counterfeit detection and defect detection and localization. THz defect localization has been used to mitigate nano- or micro-sized defects in the VLSI design, while counterfeit detection has focused on micro- and macrodifferences in the IC die, epoxy molding, or pins. A beginner’s level introduction to IC packaging has been compiled by Lancaster and Keswani,¹³⁷ while FA techniques for 3D packaging were reviewed by Altmann and Petzold,¹³⁸ Li et al.,¹³⁹ and De Wolf et al.¹⁴⁰ Interconnect quality and reliability for 3D IC packages were reviewed as part of Springer’s “Series in Advanced Microelectronics” book covering microstructure changes and failures driven by mechanical, electro, and thermal stresses, which are important as many modern mobile devices fail due to high frequency impacts with the ground.⁸⁶

Although there are many non-THz-based inspection methods for ICs, it is difficult to reduce the processing time of assessment and interpretation of signals for many IC NDT. These issues should be addressed for the development of an effective in-line inspection system.⁶⁶ Aryan et al. provided a review of the state-of-art, as of 2018, NDT methods used for evaluation of IC packaging with a critical conclusive findings:

5.2.

Electrical Defect Localization

5.2.1.

Time-domain reflectance

Time-domain reflectance or TDR matured into an NDT fault location tool, utilizing powerful pulses, of extended electrical systems such as wiring or cables. With increased signal processing, the application was extrapolated to FA of semiconductor components such as board, circuitry, and package interconnections. TDR utilizes a connected medium such as copper wire while transmission methods such as x-ray or acoustic imaging propagating through the medium of air. TDR is possible through monitoring of electrical pulses which are sent through the circuitry of a device, and the reflection is monitored for irregularities such as shorts or disconnections. The resulting time-domain waveform obtained in TDR analysis is determined by both the input pulse waveform and transfer function of input point. In general, if an IC chip is not powered, it will become a linear time invariant system.¹⁴¹ Due to increased packaging complexity for ICs, the determination of a fault was possible using TDR, but the localization of the defect’s exact location became difficult due to poor spatial resolution. Ultrafast THz pulses were used in combination with TDR to resolve failure points with a resolution surpassing conventional electrical TDR method alone.¹⁴² Nondestructive FA of advanced 3D packaging still became more challenging to perform due to the much smaller interconnects such as TSV and microbump for both electrical and physical inspection methods even for ultrasfast laser pulses. For accurate analysis, there are two major challenges in realizing a THz-TDR system. First, the generated THz impulse signal must be coupled into the circuity with a minimized loss; second, the reflected THz impulse signal must be coherently detected in time domain.¹⁴³ Frequencies above 110 GHz are not transmitted well by electrical contacts used for TDR, and hence most of the resolution power is lost within the TDR system during reflection. Nagel et al.¹⁴¹ addressed these coupling and detection challenges and enabled high spatial resolution noncontact electro optical TDR inspection by incorporating a TDS system and custom-built probes with a tip radius of $0.2\mu \mathrm{m}$. This THz-TDR instrument could locate discontinuities in electronic structures with a spatial resolution of $0.55\mu \mathrm{m}$ through near-field collection in proximity to the device via probe/wave guide coupling. Nagel et al. demonstrated successful microprobe analysis of a device under test DUT structure: a typical Cu-based coplanar waveguide with an electrode width of $50\mu \mathrm{m}$, $30\mu \mathrm{m}$ spacing, and $3\mu \mathrm{m}$ copper thickness based on an undoped Si substrate.¹⁴⁴ A 75-nm-thick layer of ${\mathrm{SiO}}_2$ is deposited between the electrodes and substrate. This method is very effective for determination of open-end, full short-cut, and semi- short-cut defect via analysis of the peak or trough of the TDR signal after probe coupling. This setup can be seen in Fig. 14(a) where an open end, full short cut, and a semishort cut are all characterized by their TDR amplitude waveforms.

Fig. 14

Left: (a) Differential time-domain reflection signals from an open-end, (b) a full short-cut, and (c) a sequence of semi and full short-cut CPW. Corresponding microscope images on the right.¹⁴⁴ Copyright © 2014 IEEE. Reprinted with permission from IEEE Access. Right: (a) EOTPR waveforms and schematic diagram of the defective circuit from case study III. (b) Optical images obtained after cross section of the C4 bump in the good and failed units. The failed unit has a rupture in the under bump metallization which is the cause of the electrical open failure.¹⁴⁵ Adapted from Copyright © 2013 International Microelectronics Assembly and Packaging Society; European Microelectronics Packaging Conference.

5.3.

Terahertz Signal Emission

5.3.1.

Signals emitted under test

In 2013, a method of THz-based “chipless” RFID tagging using an IC’s dielectric features as opposed to the information encoded in a barcode or qr code tag,¹⁵⁷ an RFID tag is compared to a THz tag in Fig. 15. In 2016, a unclonable chipless tag based on concentric ring slot resonators was presented that utilizes manufacturing variations, without additional hardware such as an RF tag.¹⁵⁸ Ali et al. presented a comparison of RF and THz emission in 2017, authentication methods based on randomness inhering in the fabrication process showing that it is possible to discriminate RF, respectively, THz, signatures obtained from devices whose differences of geometrical parameters are as small as tens of micrometers.¹⁵⁹ Ali in 2019 reviewed chipless tags identifying issues with cloning due to the use of natural randomness in the fabrication process and their use for tagless authentication.¹⁶⁰

Fig. 15

(a) RF tag based on RF waves which can be read remotely without direct human interaction between the reader and the tags, (b) THz tag where “chipless” tags are most often made of a dielectric substrate on which a conductive pattern is deposited. The specific geometry allows the tag to be identified with certainty.¹⁵⁷ Copyright © 2013 IEEE. Reprinted with permission from IEEE Proceedings.

Ahi et al. developed an engineered nanostructure for tagging and protection of an IC, where numerical matrixes on an IC surface illuminated by a laser creating an encrypted tag that is difficult to duplicate;^161,162 however, this requires expensive lithography equipment and increases the cost of production for each chip during packaging. Lower cost methods of THz tagging were proposed in 2017, where a low-cost photopolymer SU-8 was demonstrated as an effective candidate for security applications because UV exposure of the film was correlated to a change in the THz frequency.¹⁶³ This change in polymer could be detected by THz while not seen optically, demonstrating a possible use for detecting the level of cross-linking in an IC package. Similar antitamper mechanisms were proposed in 2020 by Cheng et al., where nanoscale strain engineering is utilized to create a self-erasable and rewritable platform for antitamper hardware based on the reversible structural changes. The changes between trans and cis isomers in azobenzene (A3) molecules combined with induced strain in the overlying tungsten diselenide (${\mathrm{WSe}}_2$) monolayer, thereby affecting its optical bandgap enabling tags to be imprinted but disappear after a set period of time.¹⁶⁴

The Naval Surface Warfare Center-Crane Division patented a method of detecting counterfeit ICs based on their emission characteristics in 2017. This new approach utilized power-pin-based characterization of capacitive and inductive signatures across a specific frequency range using a precision impedance analyzer.¹⁶⁵ Manufacturing variation utilizes natural variations during fabrication specific to each facilities or foundry, and two manufacturers even with identical tools and processes cannot fabricate an identical die. Electrical signals are connected and sent through an IC’s power-pin while characterizing the distortion field created by the IC’s internal capacitance and inductance. Through sweeping the electrical signal input across a specified frequency range, a fingerprint specific to the manufacturer can be derived, while returning minute deviations from expected to accurately identify an IC die as being fabricated legitimately or maliciously. Authentication methods based on THz tagging have been used extensively in 2018 and 2019 to create automated techniques,^{157,166–169} along with methods of detecting spectral fingerprints from THz-tags behind shielded materials using machine learning.¹⁷⁰ Similar tagging methods have been combined with physically unclonable functions PUFs to create unclonable tags for protecting counterfeit pharmaceuticals, an industry plagued by counterfeits similar to semiconductors.¹⁷¹

Laser scanning or probing of IC designs such as VLSI or monolithic integrated circuits is possible with EOTPR; however, it is limited by SNR of the laser probe and reflected pulses. Tag-based methods require a modification of the IC package during fabrication to allow for chipless tagging; however, methods based on manufacturing process variations have been proposed in 2020.¹⁵⁹ A new approach for THz signal emission-based inspection by Shur et al. is based on detecting the electrical response generated at an IC’s external pins while the IC is irradiated by THz or sub-THz radiation.^172,173 As opposed to EOTPR methods where a reference signal or golden sample is required for comparison of expected to defective signals, this method is able to become self-learning and self-improving through AI and ML algorithms. A method building on Shur’s work incorporated AI techniques for ensuring hardware security by obtaining images that depend on the THz beam’s radiation polarization, frequency, impinging angle, and depth of focus, thus allowing to record unique and detailed information about the chip that could be used for an etalon image to identify (by the image comparison) faked chips, chip defects, and even to predict reliability and lifetime.¹⁷⁴ EOTPR is limited by being a comparison-based method, often requiring a KGD, and for this reason methods that incorporate verification without a golden sample are of interest. Tagging methods are limited by manipulation, added cost, and environmental conditions, and it is important to create a signature for a device based on its physical characteristics using THz imaging in transmission and reflectance. Image-based techniques are able to characterize an IC based on the inherent structure without relying on a golden sample or complex tag.

5.4.

THz Imaging-Based Methods

5.4.1.

THz imaging generation

The use of THz radiation to form images has been used for IC packaging inspection since 1995,⁹ shown in Fig. 16 due to its ability to penetrate the optically opaque epoxy molding with a spatial resolution of $250\mu \mathrm{m}$. The THz range contains information from across multiple inspection modalities, such as microwaves, infrared, visible, ultraviolet, or x-ray images. This combination of specular and diffuse features for THz is shown in Fig. 17 where THz spectrum combined the effects of microwaves, visible, and IR wavelengths. Combining of these various modalities can be represented with a THz image; and compared with images formed using lower frequencies such as infrared, THz images have the advantage of superior spatial resolution due to its shorter wavelength.⁵⁹

Fig. 16

(a) A THz image of a semiconductor integrated-circuit chip package. The THz image collected using a near-field scanning system TeraCube from Protemics used to collect high-resolution Terahertz data with a photo-conductive microprobe. The THz transmission is color coded according to the total intensity of the transmitted THz signal isolated over a 1.996-THz frequency range. The plastic packaging material of the chip shows only little absorption in the THz spectral range, whereas metals are fully absorbing and doped semiconductors are partially absorbing.

Fig. 17

Surface scattering across the EM spectrum. (a) Most surfaces appear smooth at microwave frequencies (specular scattering), (b) same surfaces exhibit significant considerable roughness in optical spectrum (diffuse scattering), (c) in THz regime most building surfaces exhibit significant diffuse scattering and strong specular reflections.⁹⁰ Copyright © 2014 IEEE. Reprinted with permission from IEEE Access.

The packaging materials of the ICs are compositions of different materials, and these materials are transparent to the laser in the THz region. By recording the reflected or transmitted beams in certain step intervals and mapping the reflected or transmitted beams on a 2D plane, analyzing an entire layer is possible on one graph. THz-TDS systems can record the intensity and time delay of the traversed THz pulse and generate the THz images.⁹²

Although, THz wavelength is more generous in size than the features identified, the use of image restoration and incorporating the THz beam’s PSF into the reconstruction increased spatial resolution to $\sim 0.1\mathrm{mm}$. With PSF and other different THz resolution enhancement techniques (RET), the imaging contract and resolution are high enough to detect the structure and material difference between genuine and counterfeit ICs. Ahi et al. used THz-TDS methods combined with PSF-based image restoration techniques to characterize and identify quality control issues that can be used for authentication of packaged ICs below the diffraction limit of the THz wavelength.^92,175 The complex internal structure and precise bond-wires of ICs are much smaller in feature size than those of external pins or packaging. These small features below the diffraction limit must be resolved to adequately characterize IC functionality.

5.4.2.

Image enhancement

RET for improving the spatial resolution of imaging has been used to resolve features below the diffraction limit of THz radiation. The THz beam can be modeled utilizing information about the sample to enable an improvement of the resolution in spatial and spectral resolution. Algorithmic techniques for resolution enhancement are often based on a deconvolution process, which depends on accurately estimating the PSF.³¹ In 2021, Zhang et al. presented a comprehensive analysis of the history of THz resolution enhancement as part of their research into an advanced THz resolution technique, with an in depth analysis relating current research to the historical advancements. The resolution enhancement for THz imaging is critical for THz techniques where 3D reconstruction is performed on THz data. Digital holography, CT, or ptychographic reconstruction methods all rely on high quality THz data collection before 3D reconstruction, and thus most reconstruction methods incorporate resolution enhancement. Similarly, for THz machine learning-based analysis methods, these incorporate a THz resolution improvement before machine learning operations or utilize machine learning to connect features and weights between modeled and empirical THz data. In 2010, Li et al.¹⁷⁶ introduced the Richardson–Lucy algorithm as a promising deconvolution approach to improving image details. The intensity distribution function of an actual THz laser focal point can be used as an approximation in the PSF in the restoration algorithm. In 2020, Li et al. resolution enhancement has been combined with machine learning. This is accomplished by integrating the focused THz beam distribution, which determines the relationship between the imaging range and the corresponding image restoration level; second, an adjustable CNN is introduced to cope with this range-dependent super-resolution problem.¹⁷⁷ This enables simple adjustment of the estimation parameters, providing a way for appropriate weights to be associated between fixed levels without additional training of the network.

Ding et al.¹⁷⁸ proposed the Richardson–Lucy and knife-edge methods to restore the scanning image. A CW THz source was utilized to characterize the intensity distribution of a sample. CW THz sources have an advantage over pulsed systems for phase retrieval of samples through transmission or reflection imaging. In 2018, THz ptychography was performed at THz frequencies and then reconstruction was used to derive the amplitude and phase with both simulated and real data with a lateral resolution accounts to $<2Å$.¹⁷⁹ In 2020, Rong et al. performed THz ptychographic imaging and resolution enhancement of a CW THz source’s diffraction patterns via a tilted plane correction. This technique can be used to retrieve the complex-valued object function while suppressing the negative effect of nonuniform illumination.¹⁸⁰ If CW THz sources used for pychtography currently can be made more portable, an in-situ nondestructive THz radiation topographic metrology would become possible.

Xu et al.¹⁸¹ presented a high-resolution reconstruction model and algorithms for THz imaging using projection onto a convex set, iterative backprojection, Richardson–Lucy iteration, and 2D wavelet decomposition reconstruction. An automated THz spectral analysis algorithm faces difficulties due to the low-resolution THz spectrum image and results in an ineffective recognition of an analyzed sample’s distinct features. In 2020, a spectrum clustering recognition model based on t-distribution stochastic neighborhood embedding was proposed to address this ineffective sample defect recognition.¹⁸² This methodology is different from the traditional method to identify target defects through images, instead a resolution enhancement method is used to identify target defects via spectral clustering, which is an important auxiliary means to identify target defects through THz images.¹⁸²

Ning et al.¹⁸³ used a constrained least squares deconvolution algorithm and eliminated the PSF physically by measuring the sample of a 0.5-mm pinhole on the THz imaging system. The postprocessing of the experimental data, given for example a priori knowledge of the system 3D PSF applying computational imaging techniques, will also increase resolution.¹⁸⁴ Many THz inspection systems such as those used for NDR are focused on deriving accurate 3D volumes without noise, and a common issue is a slow imaging time or scanning rate. Advanced methods require lengthy processing time per pixel and perform raster scanning to collect high-resolution THz data from across a sample.

Although the THz beam can be accurately modeling and matches experimental data, the simulated or modeled PSF cannot show the attenuation of the beam in the object. In addition, the diameter of the pinhole increases the uncertainty of the PSF.¹⁸⁵ Ahi³¹ mathematically modeled the transmission PSF of the THz imaging systems within the scale of a Gaussian beam to enhance the resolution of the THz imaging. As a result of implementing this RET, the accuracy of the measurements on THz images has been improved from 2.4 to 0.1 mm and bond wires as small as $550\mu \mathrm{m}$ inside the packaging of the ICs are imaged. Bond wires are the intermediate-sized features in the IC package between the nanoscale features of the die and the macroscale pins and external connections.

In theory, the reflected THz signal is the convolution of the incident THz pulse, which should ideally consist of time-shifted ideal impulses corresponding to the material structure. The Fourier transform of this ideal impulse response function is broadband and represents the complex components of the sample through which the pulse has passed through across the entire frequency spectrum. Conventional deconvolution based on direct inverse filtering aims at retrieving the impulse response function by applying the inverse Fourier transform of the transfer function.¹⁸⁶

THz systems have many sources of noise that can cause the generation and detection to deviate from modeling the PSF through deconvolution. For pulsed system specifically, there is fluctuation noise randomly generated by the ultrafast laser pulses. In addition, there is the shot noise in the THz detector and the thermal radiation noise generated during the running of the THz system.¹⁸⁷

Therefore, it is important to model the impact of noise on THz signals when enhancing the resolution of the THz imaging outside of using deconvolution techniques alone. Ahi³¹ denoised the THz signal using the fast Fourier transform (FFT), by designing a filter (low-pass, high-pass, or band-pass) to block the frequency band where the noise is located. However, such FFT-based methods typically sacrifice spatial resolution due to the lack of time-frequency localization,¹⁸⁸ and they cannot effectively separate high-frequency detail from noise when denoising nonstationary signals. This can result in oversmoothing owing to the loss of useful information.¹⁸⁹

Mittleman et al.¹⁰ first proposed using a wavelet transform for THz signal processing, because the form of a THz pulse is similar to the common wavelet basis functions.¹⁹⁰ Advanced methods in 2020 utilize autoregressive spectral extrapolation (AR) based on the modified covariance method (AR/MCM) and have compared it to Wiener filtering combined with wavelet denoising. AR/MCM does not discard any frequency components in the low signal-to-noise regions of the measured data, which can reduce noise but cannot effectively reconstruct information outside the high-SNR band.¹⁹¹

Due to the characteristics of time-frequency localization found using variable size time-windows for different frequency bands,¹⁹² the wavelet transform provides a powerful tool for improving SNR of the THz signal. In 2020, Zhang et al. proposed a method for THz image restoration using wavelet denoising with PSF model that enhanced the resolution and quality on the imaging detection of the IC chips with complex internal structures.³⁴

Ahi et al. used THz-TDS methods combined with PSF-based image restoration techniques to characterize and identify quality control issues that can be used for authentication of packaged ICs^30,193 shown in Fig. 18. This method was able to detect the presence of unexpected materials in counterfeit devices, blacktopping layers (used by counterfeiters to hide the original label and overprint a false one), shape and dimensions of hidden structures, sanded and contaminated devices, differences between internal structures of counterfeit and authentic devices, such as misshapen die-frames and bond-wires.⁹² In 2019, a review of polymer interactions frequency ranges from 0.5 to 5.0 THz were measured in several epoxy resin nanocomposites with and without coaddition of microsized fillers, aiming at obtaining important information on intermolecular interactions between filler and polymer.¹⁹⁴ The characterization of ICs and their metallic and polymeric composition can be performed at high speed through transmission or reflectance mode THz imaging methods.

Fig. 18

A resolution enhancement process for THz imaging with enhanced resolution can be implemented in parallel with optics-based RETs and add to their effectiveness. This method is implemented as a system comprised of various resolution and quality enhancement techniques. In this system, filters in time and frequency domains are used to filter out the noise, low-frequency spectrum, and diffraction distortions. Reprinted with permission from Ref. 193, Copyright © 2019 Elsevier.

5.4.3.

Transmission THz imaging

The IC structure is comprised of many different materials from plastic, ceramic, to metals. This variation in composition resulted in a differential attenuation during transmission-based inspection methods such as x-ray or THz imaging. This contrast is due to how each material absorbs x-ray or THz radiation, which enables distinguishing of internal features such as number of bond wires, die shape, or different types of polymers or fillers. THz imaging can resolve more spectral and depth information than x-ray, but is limited by highly attenuating features or samples. THz-TDS has the ability to correlate the time delay of a THz signal with different amplitudes can be used to isolate features layer by layer without the need for CT, THz-TDS reflected or transmitted pulses can represent layer thickness of varying material inside the IC packaging. This means THz can be used to nondestructively extract the interstructure of ICs in transmission similar to x-ray imaging without the need for multiple angles of collection to reconstruct the inner 3D structure. The direct relationship between IC THz imaging and x-ray imaging has been shown by Ahi et al.⁹² and can be used to increase the resolution of the THz image.¹⁷⁵ However, the image quality of THz imaging is lower than x-ray imaging, but THz systems can operate at much higher collection speeds while offering a lower risk of operation for employees due to nonionizing radiation. For the IC counterfeit detection and quality control, THz imaging spatial resolution has been demonstrated as adequate for counterfeit identification (Figs. 19 and 20).

Fig. 19

Images of (a)–(f) an authentic IC and (g)–(l) counterfeit using various algorithmic reconstruction methods.¹⁹⁵

Fig. 20

THz reflection images on the left and optical images using a high-resolution microscope on the right. (a) An authentic IC. (b) A counterfeit recycled IC: the contaminated spot is obvious in the THz image. (c) Image of an IC which is sanded on one side. Reprinted with permission from Ref. 185, Copyright © 2018 Elsevier.

THz nonionizing radiation offers more advantages for characterizing advanced packaging where low-k material is used as an IC substrate, and if inspected with excessive ionizing radiation such as x-ray, it can become irreversibly damaged. The THz laser source can be used to inspect these devices without any damage. Second, for the x-ray, the input laser will traverse through the packaged IC laterally, which will cause the attenuation of out-signal. This signal attenuation will cause the inaccuracy of some characterization, such as thickness measurement. However, for the THz imaging, the input signal will be perpendicular to the IC packaging, which has much lower signal attenuation. This will increase the accuracy of measurement, and for some broad packaging such as Quad Flat Package, this accuracy of THz measurement will be more attractive.¹⁸⁵

6.1.

Critical Dimension Metrology

The semiconductor industry is running out of metrology tools that are NDT while able to characterize atomic level deviations shown in Fig. 21. Electro-optical frequency mapping (EOFM) is responsive to carrier densities in electronic devices. EOFM demonstrates promising results for a characterization of vertical high-performance. (HBT) in SiGe:C BiCMOS by demonstrating distinguishable results of a golden and a faulty SiGe:C HBT.¹⁹⁶ The semiconductor industry is testing new metrology methods capable of meeting the future requirements to characterize advanced 3D transistor structures shown in Fig. 22 where the CDs are <10 nm. X-ray scattering techniques are one candidate owing to the subwavelengths which are sensitive to internal changes in electron density.¹⁹⁷ The semiconductor and nanotechnology industry anticipates facing future difficulties with quantitatively characterizing nanoscale patterns using conventional microscopy techniques, such as SEM, AFM, or optical scatterometry.¹⁹⁸ New methods for CD measurements are required to expand the NDT options for needed characterization of structures during fabrication and FA workflows. Postsilicon testing using dynamic laser-based methods is increasing in adoption as it can identify defects in sub-20-nm FinFET technology nodes.¹⁹⁹ Intense THz fields interacting with n-type InSb semiconductors has shown that as the THz field strength reaches to $1\mathrm{MV}/\mathrm{cm}$, its application for material characterization becomes considerable.²⁰⁰

Fig. 21

Schematic demonstration of the technology gap between the IC package size and the spatial resolution of NDT methods.⁶⁶

Fig. 22

IRDS roadmap showing the scaling of standard cell height and width through fin depopulation and device stacking, The IRDS more Moore roadmap details the challenges of realizing the required levels of reliability, which (1) scaling, (2) new materials and devices, (3) more demanding mission profiles (higher temperatures, extreme lifetimes, high currents), and (4) increasing constraints of time and money. These reliability challenges will be exacerbated by the need to introduce multiple major technology changes in a brief period of time. Figure reprinted from The International Roadmap For Devices And Systems © 2020 IEEE.

6.2.

THz Conductivity

Silicon wafers dimensional accuracy from polishing and conductivity from doping are critical features measured throughout the fabrication process of an IC. THz time of flight systems such as TDS have been used to determine wafer thickness after polishing via an NDT.²⁰¹ Thickness of the wafer affects electrical and structural properties, and noncontact methods enable in-line testing during fabrication.

THz-TDS has been successfully used to characterize the charge carrier density in silicon wafers showing that varying the illumination intensity affects the THz-TDS signal, because of the change of charge carrier density. THz-TDS silicon wafer measurements were compared the microwave photoconductive decay method (MWPCD) demonstrating that defective areas identified with the THz measurements coincide with those identified by the MWPCD method.²⁰² THz is therefore comparable but is superior as it is an NDT method for wafer analysis as it is able to identify mechanical and electrical properties while MWPCD requires contact with the sample and only characterizes conductivity.

Since 1998, near-field imaging with ultrashort, broadband far-infrared pulses has been of interest for the use ever since demonstration of a resolution below THz wavelength (Lambda/4) upon a “resolution test pattern” demonstrating spatial resolution is limited by the size of the near-field tip’s aperture.²⁰³ In 2001, Hillenbrand et al. advanced scattering-type or apertureless SNOM to a spatial resolution of 1 and 10 nm, of gold upon silicon substrate, emphasizing the importance of future work into elimination of artifacts and the understanding of near-field contras in s-SNOM imaging.²⁰⁴

Adam’s review of near-field THz measurements methods and their applications in 2011 give an overview of the history of the field to describe the different approaches, applications, and perspectives for near-field.²⁰⁵ Since 2011, there has been extensive developments in the spatial resolution and capabilities of the THz near field. Keiser et al. presented advances in THz science and technology that rely on confining the energy of incident THz radiation to small, very subwavelength-sized regions with a focus on two broad areas of application for such field confinement: metamaterial-based nonlinear THz devices and THz near-field microscopy and spectroscopy techniques.²⁰⁶ Wang et al. reviewed the field of THz modulators based on different kinds of dynamic tunable metasurfaces combined with semiconductors, 2D electron gas heterostructures, superconductors, phase-transition materials, graphene, and other 2D material.²⁰⁷ Graphene is a critical material related to the success of THz modulation due to how graphene plasmon resonances have remarkably large oscillator strengths, resulting in prominent room-temperature optical absorption peaks compared to the low temperature requirements for other 2D electron gases. In 2011, Ju et al. developed a technique that enabled the fine control of graphene plasmon resonances over a broad THz frequency range by changing microribbon width and in situ electrostatic doping opening the door for widescale research into potential graphene-based THz metamaterials.²⁰⁸

Pala and Abbas²⁰⁹ reviewed the use of LTEM THz imaging at the nanoscale for VLSI and transistors. In the same year, Gu et al.²¹⁰ integrated photoconductive silicon into the metamaterial unit cell, causing switching of the transparency window under excitation of ultrafast optical pulses such as THz, allowing for an optically tunable group delay of the THz light. This opened the possibility for designing chip-scale ultrafast devices for optical buffering and THz active filtering. Thereafter, graphene field effect transistors (FET) were utilized as an effect method of THz detection bringing THz detection to realistic settings, enabling large-area, fast imaging of macroscopic samples.²¹¹ Ferrari et al.²¹² reviewed the potential roadmap for graphene, 2D crystals, and hybrid systems detailing that graphene can exceed typical room temperature THz detection limits by exploiting THz plasma waves that are weakly damped in high-quality samples, allowing for resonant detection regimes in an FET.

Joyce et al.²¹³ reviewed THz conductivity spectroscopy for silicon NWs to great success. Kužel and Němec²¹⁴ reviewed THz spectroscopy of nanomaterials in 2019 identifying key advantages of THz conductivity analysis:

In 2019, THz microscopy and applications in semiconductor testing was presented by Xu et al., reviewing THz microscopy and detailing commercially available equipment that can enable near-field THz imaging from Protemics, NeaSpec, and Teraview.²¹⁵

6.3.

Near Field (LTEM, S-SNOM, THz-AFM)

THZ-TDS pulsed methods have advanced in spatial and spectral resolution enabling smaller feature identification using the THz wavelength. These advances have resulted in the successful application of THz radiation to measure charge carrier dynamics in semiconductors materials. Due to the low energy of THz radiation (10s to 100 meV), THz wavelength responds to common semiconductor dynamics such as those of charged quasiparticles. Each quasiparticle such as free carrier has an identifiable fingerprint in the THz region providing a way to characterize the ultrafast physical electrical properties of semiconductor materials. This frequency region is also well suited for the study of phonon resonances and intraband transitions in low-dimensional systems. The nonequilibrium time evolution of carriers and low-energy excitations is critical to the understanding of how transistors and gates operate, and THz sources and detectors enable a time step not achievable with traditional optical or e-beam methods directly, a sub-ps time resolution. Being an all-optical technique, THz time-domain spectroscopy is contact-free and noninvasive and hence suited to probe the conductivity of, particularly, nanostructured materials that are difficult or impossible to access with other methods.⁵⁴ Near-field TDS methods can enable rapid chemical spectroscopy and penetration beneath substrates such as silicon. THz-TDS methods are able to detect a sample’s spectral features using THz pulses and the known time delay of the pulsed signal, while LTEM methods are based upon a sample’s own generation of THz radiation such as photoconductive antenna of semiconductor components at the nanoscale. This emission of THz radiation is very low power but can be detected by probes near the surface, making this method exclusively near-field. At present, it is unclear which are the best measurement methods needed to evaluate the nanometer-scale features of semiconductor devices and how the fundamental limits of physics will affect the required metrology.²¹⁷ S-SNOM systems offer the benefits of multiple tools such as AFM, LTEM, THz-TDI, and optical-pump THz-probe in the same configuration providing a spatial resolution on the nanoscale.²¹⁸

6.4.

THz-AFM

Scanning microwave microscopy is based on similar principles to THz near field, imaging. SMM combined with THz has been used to image nanostructures 4 to 15 nm below a surface with 30 to 50 nm spatial resolution.²¹⁹ This technique shows promise for an NDT of characterizing doping concentration of source drain regions along with dimensional analysis of transistor structures. THz near-field techniques have been combined with AFM to characterize surfaces with high detail; however, AFM requires contact with a surface and combined THz AFM systems are considered a destructive test method. Low cost near-field semiconductor analysis is a product of commercially available equipment such as PCAs can be used to generate a wide bandwidth pulse with frequencies from 0.1 to 3 THz. These PCAs pulses are then coupled with a probe consisting of metal, commercially available as an AFM tip. The probe size limits the resolution, and its size is specifically fabricated to create an antenna coupling with the peak wavelength of the incoming THz pulse. THz near-field signals are scattered by the metallic tip into the far-field region where these signals can be collected and focused onto the second PCA for detection. Due to the THz interactions with the metallic tip, only the frequencies up to 2 THz are efficiently scattered while the rest are absorbed. This limits the THz bandwidth of the near-field signal to below that of the PCA. The detection scheme for THz-AFM is identical to conventional THz-TDS, except THz signals are demodulated at higher order harmonics of the tapping frequency of the AFM tip to isolate the near-field signal of the scattered light as opposed to the large background noise present during collection.²²⁰

6.5.

LTEM

Non-THz-based optical methods such as near-field scanning optical microscopy (NSOM) of semiconductor materials have been used to obtain subwavelength resolution of surface features.²²¹ However, this NSOM is only applicable to characterization of defects visible in IR or visible light not obscured by connections often found in the complex BEOL and FEOL structures.

Kawase presented LTEM imaging of VLSI noting that in the case of multilayered circuits, rear illumination through the substrate semiconductor of the chip is necessary because the multilevel interconnections in the chip prevent the generation and detection of THz emission.²²²

THz LTEM methods operate upon femtosecond lasers that can excite electrical structures such as transistors and circuitry and detect the emitted THz radiation for characterization. When the laser is incident on photoconductive structures, it generates a transient photocurrent, causing a THz pulse to be emitted. Structures capable of emitting THz, and therefore suitable to be examined by the LTEM, include photoconductive switches with an external bias voltage (i.e., Auston-type switches), unbiased interfaces carrying electric fields (e.g., p-n junctions, Schottky contacts), and some semiconductor surfaces.¹ LTEM transmission measurements on an IC using SIL, with spatial resolution of $1.5\mu \mathrm{M}$ were performed in 2008, an example of this LTEM imaging can be seen in Fig. 23.²²³

Fig. 23

LTEM images of LSI fabricated by 180 nm process (a) without and (b) with a defect. The localization of LSI defects is one of the key issues for the current and future LSI developments.²⁵ CMOS indicated by the red arrow in (b) includes a signal line disconnected artificially while fabricating, which makes the LTEM amplitude smaller, and the corresponding CMOS image becomes darker. (c) Note that the fs laser pulses are irradiated from the back side and the THz signals are detected from the back side because the front side is covered by metal layers. A GaP solid immersion lens is attached to the back side of the LSI, to improve the spatial resolution.²²³ The whole LTEM image size is $30\times 30{\mathrm{mm}}^2$, and a resolution estimated by the transition width from 20% to 80% of the amplitude is 360 nm as indicated in (d). Caption adapted from Mittleman.⁵⁹

Yamashita et al. developed and refined the hardware for an LTEM system, achieving a spatial resolution of a few $\mu \mathrm{m}$ in Refs. 19, 25, 224225.226.–227. The progressive developments improved the spatial resolution of the technique while extending it to the backside of the device where THz signals in the near-field are captured without transmitted through the metal layers.²⁵ The contrast mechanism of the LTEM imaging is based on the THz field amplitude and its proportion to the local electric field or the background noise. The amplitude can be plotted after a raster scan of the surface and used to develop an image of areas where THz emission is most intense. The capabilities of the LTEM were improved to a spatial resolution of $0.6\mu \mathrm{m}$ by incorporating near-field techniques.²²⁸ A related method was used to perform time-resolved transmission imaging, demonstrating measurements of carrier lifetimes and mobilities with a spatial resolution of $60\mu \mathrm{m}$.²²⁹ Recent advances in LTEM and s-SNOM were compared by imaging a lightly p-doped InAs sample, and we were able to record waveforms with detector signal components demodulated up to the 6th and the 10th harmonic of the tip oscillation frequency and measure a THz near-field confinement down to 11 nm.²³⁰

6.6.

S-SNOM

Due to the nanoscale features of a transistors, an optical spatial resolution of $<100\mathrm{nm}$ is necessary for transistor characterization. Huber et al. demonstrated an SNOM technique based on single frequency enables higher spectral resolution, a 2.54-THz gas laser, achieving a spatial resolution of under 50 nm.²³¹ This technique of THz-SNOM imaging was able to detect individual transistors and their material composition while deriving electron carrier concentration. S-SNOM systems are currently too sophisticated for in-line monitoring of devices but can provide off-line characterization for FA. S-SNOM is based on a metal probe tip that scatters an incident beam, such as THZ, where the radiation becomes enhanced in the nearby area to the tip. The contrast in s-SNOM spectroscopy is based on dipole coupling between the probe tip and surface, which the metal probe is sensitive to the dielectric constant of the surface.²³² By introduction of an optical gating beam on a semiconductor wafer, near-field THz imaging with a dynamic aperture has been realized. The spatial resolution is determined by the focus size of the optical gating bean and the near-field diffraction effect. THz imaging with subwavelength spatial resolution (better than $50\mu \mathrm{m}$) is demonstrated. Especially, with the recent closing of the “THz gap” in s-SNOM operation, this nanoscope is well capable of quantifying the conductivity of state-of-the-art nanoelectronic devices, as well as for exploring exotic conduction mechanisms provided by heavy, correlated, Cooper-paired, or spin-aligned electrons.²³³ Recent adaptation of s-SNOM has been used to characterize NWs,^42,234 2D materials, and nanoparticles. An s-SNOM method characterized the axial carrier density gradients in phosphorus-doped silicon NWs by quantitatively determining the carrier density and length of the doped segment as well as the functional form of the charge carrier gradient in the transition region between doped and nominally undoped segments.²³⁵ THz nanoimaging of exfoliated single and multilayer graphene flakes has been demonstrated using a state-of-the-art (s-SNOM) system.²³⁶ The interrogation depth or profiling depth of s-SNOM is a limitation in its ability to characterize structures behind layers, which currently limits the backside probing of transistors with s-SNOM. However, an s-SNOM method can sense the material contrast between gold and silicon under a PMMA layer with a thickness of larger than 100 nm and the subsurface resolution is better than 100 nm. The above studies illuminate the capability and potential application of s-SNOM in subsurface imaging and defect inspection at the nanometer-scale resolution.²³⁷ It has been shown that for metallic structures, the near-field material contrast of s-SNOM depends strongly on the size of the structures, although the spatial resolution is nearly independent of the structure size (Fig. 24).²³⁸

Fig. 24

(a) TEM image of a single transistor. The highly doped regions below the source and drain NiSi contacts are marked by dashed yellow lines. (b) Infrared image of a single transistor ($\lambda =11\mu \mathrm{m}$). (c) High-resolution THz image of a single transistor showing all essential parts of the transistor: source, drain, and gate. The THz profile extracted along the dashed white line (averaged over a width of 12 nm and normalized to the signal obtained on the metallic NiSi gate contact) allows the estimation of a spatial resolution of about 40 nm, from the strong signal change at the ${\mathrm{SiO}}_2/\mathrm{SiN}/\mathrm{NiSi}$ transition.²³¹

Continual improvements in TEM workflow automation and quality control have enabled CD-TEM to be used as a complementary near-line metrology method, allowing large-scale process monitoring of 3D structures via direct measurement for the first time.²³⁹ This process is reliant upon sample preparation and for effective characterization the sample thickness must be optimized during sectioning to isolate the feature of interest without capturing neighboring devices. CW lasers have been combined with TEM to achieve attosecond time resolution of spectroscopic features. In addition, this technique also shows how simple modifications can turn almost any electron microscope into an attosecond instrument, useful for visualizing atomic light–matter interactions. This combined method is achieved by synchronizing the optical cycles of a CW laser to bunch the electron beam inside the TEM into electron pulses that are shorter than half a cycle of light. The pulses arrive at the target at almost the full average brightness of the electron source and in synchrony to the optical cycles.²⁴⁰

EUV lithography is currently being used for HVM at the 7- and 5-nm node. This new lithography method will reduce the manufacturing complexity since it enables the return to the use of single exposure method to pattern the critical layers.²⁴¹ There are many challenges for the fabrication of transistors at this scale and complexity, which have to be dealt and solved. Radamson et al. reviewed the transition from 2D planar MOSFETs to 3D fin field effective transistors (FinFETs) and how the process flow faces different technological challenges. Issues faced by the semiconductor industry were covered such as nanoscaled patterning and process issues related to gate and (source/drain) S/D formation as well as integration of III-V materials for high carrier mobility in channel for future FinFETs.²⁴²

As more complex transistor devices reach the atomic scale, there will a requirement for extreme high spatial resolution at fast scanning speeds. 2D devices offer opportunities to surpass the Von Neumann architecture²⁴³ opening a path beyond CMOS and toward THz frequencies transistor operation; however, these new memory and logic devices offer new opportunities for malicious design at the transistor level.²⁴⁴