Enabling Applications of Electromagnetic Waves at 0.3–1.0 THz Using Silicon Electronic Integrated Circuits

Abstract

Over the past 15 years, the output power of silicon submillimeter-wave electronics has increased by a factor greater than 1000 reaching −3.9 dBm at 440 GHz for a single unit in CMOS and −10.7 dBm at 1.01 THz for a 42-element array in SiGe BiCMOS. The smallest power of a 1 kHz bandwidth signal at 420 GHz that can be detected has improved by 100 million times. These and the expected improvements from the ongoing activities should be sufficient to support high resolution imaging with a range of up to several hundred meters, gas sensing up to ∼1 THz, and communication over ∼1000 m. The silicon IC technologies enable integration of complex systems into a small form factor and reduction of manufacturing cost. When broad deployment of submillimeter wave systems for everyday life applications becomes necessary, the silicon IC infrastructure will be the most capable to support the high-volume manufacturing need.

Introduction

The electromagnetic waves with frequencies between near 0.3 and 1.0 THz belong in a portion of the spectrum that is commonly referred to as the terahertz (0.3–10 THz) region. The frequency band from 0.3 to 3 THz is also referred to as the submillimeter-wave (SMMW) region. Being in between electrical (millimeter-wave) and optical (far-infrared) spectra, the terahertz region has long been the interest of scientific study mostly for spectroscopic experiments.¹ Both electrical and optical technologies are adopted to generate and detect terahertz signals, as will be discussed in the second section. The submillimeter electromagnetic waves have a variety of applications,¹⁻⁷ including those that can be potentially used in everyday life.⁸⁻¹⁰ The waves can be used for real-time imaging, including 3D through walls for locating wires and pipes; through decorative pieces to detect air voids and cracks; and through low visibility weather conditions like fog, rain, smoke, dust, and others.^2,3,10 These waves can also be utilized for electronic smelling using rotational spectroscopy to detect harmful gases, for breath analysis to monitor health, and more.⁴ Additionally, the wide bandwidth available in this part of the spectrum can be used for high data-rate wireline⁵ and wireless^6,7 communication to help meet the ever-increasing demand for bandwidth in the connected world. The SMMW wireless communication can also be more energy efficient.⁹ Within the SMMW region, the 0.3–1.0 THz range is where silicon electronic integrated circuits (Si IC) can play critical roles, since both the performance and the integration level of Si IC’s have reached the point where capable electronic SMMW systems can be affordably realized.

Recent advances of silicon integrated circuits technologies, including SiGe-HBT’s (hetero-junction bipolar transistors),¹¹⁻¹³ CMOS (complementary metal oxide semiconductor),^14,15 and BiCMOS (bipolar-CMOS) IC technologies^12,13 have made them an alternate means to implement capable systems for the applications.¹⁶⁻²⁷ CMOS is the cost-effective IC technology used to manufacture the bulk of electronics including those for smartphones, tablets, PC’s, and others. The differentiating capabilities of silicon electronics for operation at 0.3–1 THz are integration of complex systems into a small form factor and reduced manufacturing cost due to the integration, high yield, and sharing of the infrastructure and cost for design and manufacturing by a large number and variety of high-volume IC products. When broad deployment of affordable SMMW electronic systems for everyday applications becomes necessary, silicon IC technology infrastructure that is presently utilized to manufacture 28 and 38 GHz 5G circuits, and 60 and 77 GHz IC’s for radar applications, will be the most suitable choice to support the high-volume and low-cost manufacturing. Lastly, experimental demonstrations suggest that the silicon IC’s can eventually be utilized to implement systems operating above 1 THz.^9,28

This paper reviews the state-of-art performance of silicon SMMW circuits, their potential applications, and expected advances that will help to expand the application opportunities. In the second section, the state-of-art performance of submillimeter wave solid-state electronics including that using III–V compound semiconductor (GaAs, InP, and GaN) electronic devices is summarized. The performance of silicon IC’s will be compared with those based on other technologies. The third section describes potential applications, examples of silicon IC’s to support them, as well future efforts that can improve the capabilities and robustness of the systems for the applications and to expand the application opportunities. Finally, the paper is summarized and perspectives on the future of silicon IC technologies for proliferation of the applications near 0.3 to 1 THz, as well as wide deployment for use in daily life, are discussed.

Performance of THz Solid-State Electronics

The basic functional blocks of SMMW systems are a transmitter and a receiver similar to those for communication systems. The same performance metrics as those for communication systems such as output power of transmitters, noise factor, F (a factor by which a signal-to-noise ratio is degraded by a receiver that determines the smallest signal the receiver can detect) of receivers, ability to handle a large range of power levels (linearity), energy efficiency, and power consumption are important. In this section, the state of the art for the two most fundamental metrics, output power and noise figure (10 log(F)), as well as the device technologies and their figures of merit that determine the output power and noise figure, are discussed.

Device Technologies for Sub-Millimeter Wave Circuits and Systems

The maximum frequency at which an active device can provide power amplification is quantified using the maximum oscillation frequency (f_max) or the frequency at which the maximum available power gain of an active device becomes unity. Another important high frequency figure of merit is the unity current gain frequency, f_T, which is a key factor determining the noise figure of a receiver. This section reviews the state of the art for electronic devices that can be used to implement SMMW circuits.

Increasing mobility, scaling-down device dimensions, and reducing parasitic loss raise f_max. As depicted in Table 1, III–V compound semiconductor (CS) materials, such as GaAs, InP, and GaN, exhibit superior electron mobility than silicon. Therefore, CS devices show a higher f_max than that for silicon devices fabricated using the same lithography capability. A 25 nm InP high electron mobility transistor (HEMT) delivers f_max of 1.5 THz and f_T of 0.61 THz, and is used to demonstrate a 9 dB gain amplifier operating at 1 THz²⁹ in 2015. 130 nm InP HBT’s support 1.15 THz f_max and f_T of 0.52 THz.^30,31 More recently, a 20 nm GaN HEMT with 550 GHz f_max and 400 GHz f_T, and a breakdown voltage of 15 V is demonstrated.³² This breakdown voltage is ∼5 times higher than that of the other III–V CS devices with comparable high frequency performance and allows the output power capability to increase by ∼25×.

Table 1. Properties of Semiconductor Materials for SMMW Devices and the Maximum Oscillation Frequency (f_max) of the Devices.

material	group	electron mobility (cm2/(V s))	bandgap (eV)	dielectric constant	fmax of transistorsa (THz)
Si	IV	1400	1.14	11.9	0.42 (NMOS), 0.72 (HBT)
GaAs	III–V	9000	1.43	12.9	1.13 (mHEMT)
InP	III–V	5400	1.34	12.4	1.5 (HEMT), 1.15 (HBT)
GaN	III–V	1500	3.4	9.7	0.55 (HEMT)

^aNMOS = N-type metal-oxide-semiconductor field effect transistor, HBT = heterojunction bipolar transistor, mHEMT = metamorphic high electron mobility transistor, and HEMT = high electron mobility transistor.

Although silicon has lower mobility, continued device scaling led to dramatic increases in f_max and f_T for the active devices in silicon IC technologies at the expense of reduced breakdown voltage and corresponding output power degradation. These have enabled demonstration of capable and useful performance for SMMW circuits and systems, though inferior compared to those of III–V CS circuits and systems. Since the CMOS technology scaling has been driven for lower DC power consumption and a higher density of digital circuits, f_max and f_T of NMOS (n-type metal-oxide-semiconductor) transistors in CMOS processes plateaued around ∼350 GHz and less than 300 GHz, respectively, when connected to the top metal level¹¹ where connections to other passive components within an IC are typically made. This occurred somewhere between the 65 and 22 nm CMOS technology nodes and makes signal generation and amplification in CMOS at frequencies above 300 GHz challenging. On the other hand, SiGe HBT technologies have been optimized for high-frequency electronics in particular to increase f_max and f_T. A 130 nm BiCMOS process offering SiGe HBT’s with f_max and f_T of 450 and 300 GHz^33,34 has been in low-volume production leading to demonstrations of high-performance circuits operating near 300 GHz.^35,36 More recently, SiGe HBT’s with f_max and f_T of 720 and 505 GHz³⁷ have been reported. Efforts to integrate such SiGe HBT’s with large-volume CMOS nodes resulted in a 130 nm BiCMOS process with f_max and f_T of 610 and 470 GHz³⁸ and a 45 nm BiCMOS process with f_max and f_T of 610 and 415 GHz.¹²

Even with circuit design techniques incorporating positive feedback, the maximum frequency of amplifiers and oscillators is limited to 60–80% of f_max.^35,39,40 To overcome this, nonlinear devices driven by lower-frequency sources are employed to produce and select desired higher-order harmonics to generate and detect signals at frequencies greater than f_max. When a nonlinear device is driven at the input frequency of f_in, signals at higher order harmonic frequencies, which is an integer multiple of f_in, are generated. By filtering out unintended harmonic signals, the output frequency can be N times greater than f_in. In the same manner, coherent detection can be performed using a local oscillator (LO) signal at a frequency N times lower than that of the input, which can be greater than the f_max of active devices used for LO signal generation. In fact, the best transmitter output power and receiver sensitivity at frequencies above ∼1 THz, and the highest operation frequency are still achieved by using nonlinear devices such as GaAs Schottky barrier diodes (SBD).⁴¹ It is also possible to realize nonlinear devices including SBD’s in silicon IC technologies. Schottky diodes and MOS varactors with a cutoff frequency of ∼2–3 THz⁴²⁻⁴⁵ have been reported and utilized for signal generation up to 1.3 THz,⁴⁶ and coherent (amplitude and phase) and incoherent (amplitude only) detection up to 1.2⁴⁷ and 10 THz,⁴⁸ respectively.

These uses of nonlinearity are similar to the generation of THz signals with a nonlinear photonic crystal utilized in photoconductive antennas (PCA) and unitraveling carrier photodiodes (UTC-PD) pumped by mode-locked lasers.⁴⁹⁻⁵² A critical difference between the photonics and integrated electronics approaches is that the drivers can be integrated with the nonlinear devices on a single platform, which reduces the form factor and power consumption owing to lower interconnect losses. The silicon IC technologies as mentioned enable large-scale integration of SMMW systems with high yield, providing a path for broad deployment.

Signal Generators and Transmitters

As discussed, transmitters are fundamental building blocks, and their output power is a key performance metric. Figure 1 depicts the maximum output power of continuous-wave room temperature solid-state electronic technologies for silicon CMOS and SiGe HBT/BiCMOS, and III–V compound semiconductors, including HEMTs, HBTs, and GaAs SBDs.⁵³Table 2 summarizes some of the key transmitter performances according to the topology and the technologies used. As mentioned, GaAs SBD circuits outperform other technologies with signal generation up to 2.7 THz with an output power of −17 and 15.4 dBm at 330 GHz.^41,54,55 For power amplifiers, InP HEMTs are used to demonstrate 4.8 dBm saturated output power (P_sat) with 0.02% power added efficiency (PAE) at 643 GHz⁵⁶ and −2.2 dBm P_sat at 847 GHz.⁵⁷ A 250 nm InP HBT is employed in an amplifier to achieve the maximum P_sat of 13.5 dBm at 301 GHz,⁵⁸ while a 130 nm technology is used to demonstrate a 620 GHz amplifier with 1.1 dBm P_sat.³¹

Figure 1.

Output power of signal generators and transmitters operating at 0.1–3 THz fabricated using silicon and III–V compound semiconductor technologies.⁵³

Table 2. Signal Generators Using Silicon and III–V Compound Semiconductor Technologies.

technologya	topology	frequency (GHz)	output powerb (dBm)	efficiencyc (%)	ref
GaAs SBD	frequency multiplier	2700	–17		(41)
	frequency multiplier	330	15.4		(54), 55
InP HEMT	amplifier	643	4.8	0.02	(56)
	amplifier	847	–2.2		(57)
InP HBT	amplifier	620	1.1		(31)
	amplifier	301	13.5		(58)
SiGe HBT	amplifier	290	5	1.2	(35)
	amplifier	255	13.5	1.47	(59)
	amplifier	305	8.3	0.7	(36)
	frequency multiplier	320	–1		(70)
	frequency multiplier	920	–17.3		(71)
	oscillator	317	–6.8	0.54	(72)
	oscillator	1000	–27	0.0073	(73)
	oscillator array	1000	–11 (42 elements)	0.0073	(73)
CMOS	amplifier	309	–8.1	1.9	(40)
	amplifier	243	10.5	2.7	(60)
	oscillator	290	–1.2	0.23	(66)
	oscillator	482	–8	0.27	(67)
	oscillator	694	–18	0.066	(68)
	oscillator array	580	0 (36 elements)	0.08	(74)
	frequency multiplier	288	0		(69)
	frequency multiplier	447	–3.2		(45)
	frequency multiplier	480	–8		(64)
	frequency multiplier	1330	–23		(46)

^aSBD = Schottky barrier diode, HEMT = high electron mobility transistor, HBT = heterojuction bipolar transistor, CMOS = complementary metal oxide semiconductor.

^bOutput power for oscillators, saturated output power (P_sat) for frequency multipliers and power amplifiers.

^cPower-added efficiency (PAE) for amplifiers and DC to RF efficiency for oscillators.

Due to the lower f_max of CMOS (∼350 GHz) transistors and SiGe HBTs (∼450 GHz) and their lower breakdown voltages, power amplifiers in silicon IC technologies exhibit 5 dBm P_sat with 1.2% PAE at 290 GHz,³⁵ 13.5 dBm P_sat with 1.47% PAE at 255 GHz⁵⁹ and 8.3 dBm P_sat with 0.7% PAE at 305 GHz³⁶ using 130 nm SiGe HBTs, and −8.1 dBm P_sat with 1.9% PAE at 309 GHz⁴⁰ and 10.5 dBm P_sat with 2.7% PAE at 243 GHz⁶⁰ using 65 nm CMOS technologies. With the advent of 720 GHz f_max SiGe HBTs, the power amplifier operating frequency is expected to increase to ∼500 GHz.

Signal generation beyond ∼300 GHz especially using CMOS is typically realized using higher-order harmonics generated by using the nonlinearity of active devices or using integrated nonlinear passive devices such as CMOS-compatible Schottky barrier diodes^48,61−63 and varactors.^45,46,64,65 The present state-of-art output power of CMOS signal generators using oscillators incorporating frequency multiplication is −1.2 dBm at 290 GHz,⁶⁶ −8 dBm at 482 GHz,⁶⁷ and −18 dBm at 694 GHz.⁶⁸ A transistor frequency multiplier is used to generate 0 dBm at 288 GHz,⁶⁹ and MOS varactor frequency multipliers are used to generate −3.2 dBm at 447 GHz,⁴⁵ −8 dBm at 480 GHz,⁶⁴ and −23 dBm at 1.33 THz.⁴⁶ SiGe HBT frequency multipliers demonstrate −1 dBm at 320 GHz,⁷⁰ and −17.3 dBm at 920 GHz,⁷¹ whereas oscillators incorporating frequency multiplication generate −6.8 dBm at 317 GHz,⁷² and −27 dBm at 1 THz.⁷³

Moreover, the output power can be increased with minimal efficiency penalty through spatial power combining by building chip-scale arrays of generators with on-chip antennas. 36 oscillators are coupled to generate a total-radiated power of 0 dBm at 587 GHz using 40 nm CMOS,⁷⁴ while the outputs of 42 oscillators are combined to generate −11 dBm at 1 THz using 130 nm BiCMOS.⁷³ In Figure 1, the maximum output power of the signal generators decreases as the operating frequency is increased with a slope of −35 dB per decade. The output power levels of silicon signal generators are ∼20 dB lower than those of III–V CS generators. By employing spatial power combining while leveraging the high-yield and integration capabilities of silicon IC processes, the gap is reduced to ∼10 dB up to 1 THz. These output power levels of silicon SMMW circuits are sufficient for many everyday applications, as will be discussed in the third section.

Incoherent and Coherent Receivers/Detectors

The second fundamental building block of SMMW systems is receivers. There are two types of signal receivers/detectors, namely, incoherent and coherent. Incoherent detectors rely on the self-mixing of incoming signals, making their design straightforward. The design can be as simple as one antenna integrated with one device with second-order (squaring) nonlinearity. However, the squaring limits its ability to detect small signals and discards phase information. On the other hand, coherent receivers/detectors mix the input with a local oscillator (LO) signal, which is stationary and whose amplitude is much greater than that of the input. As a result, orders-of-magnitude better sensitivity can be achieved while retaining the phase information for additional downstream signal processing. Integration complexity and power consumption of the LO are the main drawbacks.

Figure 2 presents a survey of the performance of incoherent and coherent detectors and low-noise amplifiers (LNA) operating at frequencies from 0.1 to 10 THz. Some of the key results are depicted in Table 3. For coherent detectors and LNAs, noise figures (NF, 10 log(F)) are used as the measure. For mixers, double-sideband (DSB) NFs are used. Noise figures directly affect the sensitivity or minimum detectible signal (MDS),

1

where k_B is the Boltzmann constant, T is the temperature in Kelvin, F is the noise factor, BW is the signal/noise bandwidth, and SNR_m is the minimum required SNR. For incoherent detectors, noise equivalent power (NEP), the input power level at which the SNR at the detector output is unity for a signal bandwidth of 1 Hz is used. MDS for incoherent detectors is also dependent on the bandwidth and SNR_m.

2

By equating the above two expressions, the equivalent F of incoherent detectors can be defined as (NEP/(k_BT√BW)).

Figure 2.

Double sideband noise figure (DSB NF (10 log(F), left) of coherent detectors and low-noise amplifiers, and noise equivalent power (NEP, right) of incoherent detectors fabricated using silicon and III–V compound semiconductor technologies. DSB NF and NEP are scaled to represent the same sensitivity for a unity signal-to-noise ratio for 1 kHz noise/signal bandwidth.⁵³

Table 3. Performance of Receivers Using Silicon and III–V Compound Semiconductor Technologies.

technologya	topology	frequency (GHz)	noise figureb (dB)	NEP (pW/√Hz)	ref
GaAs SBD	mixer	2500	15		(41)
	detector	330		1.3
InP HEMT	amplifier	850	11		(75)
SiGe HBT	detector	300		1.9	(79)
CMOS	amplifier	200	9		(76)
	mixer	260	11		(77)
	mixer	320	16.5		(17)
	mixer	425	17		(78)
	detector	10000		2000	(48), 80

^aNEP = noise equivalent power, SBD = Schottky barrier diode, HEMT = high electron mobility transistor, HBT = heterojuction bipolar transistor, CMOS = complementary metal oxide semiconductor.

^bDouble sideband (DSB) noise figure for mixers.

Similar to the output power, coherent receivers using GaAs SBDs exhibit the best performance above 1 THz with 15 dB NF at 2.5 THz.⁴¹ InP HEMT LNAs exhibit better noise performance below ∼850 GHz at which an amplifier achieves NF of 11 dB.⁷⁵ The state-of-the-art performance of circuits using NMOS transistors and SiGe HBTs is comparable up to ∼1 THz. The lowest NF for CMOS LNAs is 9 dB at 200 GHz.⁷⁶ The DSB NF for mixer and intermediate frequency amplifier combinations are 11 dB at 260 GHz,⁷⁷ 16.5 dB at 320 GHz,¹⁷ and 14 dB at 425 GHz.⁷⁸ The NF increases with a slope of ∼15 dB per decade of frequency increase. These noise figures are less than 5 dB worse compared to those of receivers using III–V receivers at the respective frequencies. Once again, as will be discussed, it should be possible to support a variety of applications using receivers with such noise figures.

The lowest reported NEP of incoherent detectors in the SMMW band is 1.3 pW/√Hz at 330 GHz implemented using a GaAs SBD and 1.9 pW/√Hz at 300 GHz implemented using a SiGe HBT in a 130 nm BiCMOS process.⁷⁹ CMOS incoherent detectors have ∼10 times higher NEP up to 800 GHz. At frequencies greater than 800 GHz, NEPs for the CMOS and III–V detectors are approximately the same. As a matter of fact, much of the results above 1 THz are for detectors fabricated in CMOS. At 10 THz, a Schottky diode detector fabricated in CMOS^48,80 achieves a responsivity (output voltage/input power) of 25 V/W. The effective noise figures of the incoherent receivers vary from 70 to 100 dB, which as expected are significantly worse than that of coherent receivers.

Applications for Silicon Integrated Sub-Millimeter Wave Systems

Silicon integrated SMMW systems have been utilized for short-range imaging¹⁹⁻²¹ through packages and fog, electronic smelling using rotational spectroscopy¹⁶⁻¹⁸ that can detect and quantify concentrations of a wide variety of gases, and high data rate energy-efficient wireless communication²²⁻²⁵ and wireline communication over 1 m dielectric waveguides^26,27 that seeks to overcome the tight tolerance limitations of optical communication systems.

Active Imaging

Silicon IC technologies have been used to demonstrate both transmission (a transmitter is located on the front side of a target while a receiver is located on the back side or vice versa) and reflection (both transmitter and receiver are located on the same side with respect to a target) mode imaging using coherent and incoherent detectors.

The imaging efforts started with transmission-mode imaging using arrays of incoherent detector pixels that integrated an antenna, a nonlinear device, and baseband circuits. A detector array composed of 3 × 5 pixels for operation at 600 GHz was demonstrated using 250 nm CMOS in 2008.⁸¹ A 1 k pixel, 0.7–1 THz array fabricated with a 65 nm CMOS process that demonstrated a 25 frame/s video capability was reported in 2012.⁸²Figure 3a shows an example of an 820 GHz imaging array using diode-connected NMOS detectors fabricated in 130 nm CMOS.⁸³ It consists of an 8 × 8 pixel array, analog multiplexers, an amplifier bank, column/row decoders, and a column of current mirrors. The average responsivity and NEP of the pixels is 2600 V/W, including the amplifier gain and 37 pW/√Hz, respectively. Active imaging systems using both a silicon transmitter and a receiver have been demonstrated up to 420 GHz using CMOS⁸⁴ and 990 GHz using SiGe BiCMOS.^72,85−89 Above 1 THz, a hybrid system that combines a high-power photonic transmitter (a quantum cascaded laser) and a silicon receiver array is demonstrated at 3.25 THz.⁹⁰

Figure 3.

(a) 8 × 8 incoherent detector array using diode-connected NMOS transistors for transmission-mode active imaging at 820 GHz. Reprinted with permission from ref (83). Copyright 2016 IEEE. (b) Fully integrated 430 GHz concurrent transceiver pixel array assembled with a 6 cm Cassegrain reflector antenna. Reprinted from ref (9) under CC BY 4.0. Modifications made. (c) Reflection-mode imaging of an object at 3 m through heavy fog using the array. Reprinted with permission from ref (21). Copyright 2022 IEEE.

As discussed in the second section, the sensitivity and SNR of the imaging system can be greatly improved by employing coherent detectors. The silicon IC technology enables on-chip integration of LO, allowing for a fully integrated coherent pixel array. A 320 GHz, 8-element array in SiGe BiCMOS,⁸⁷ a 240 GHz, 32-element array,⁹¹ and a compact pixel at 605 GHz⁹² in CMOS with improved sensitivity are demonstrated, suggesting the feasibility of coherent imaging with higher SNRs using silicon IC technologies. The SNR that can be achieved using silicon ICs is better than that of commercially available systems operating around ∼300 GHz^10,93 for inspections in laboratories, warehouses, and factories. The expected number of systems manufactured and sold, however, is not very high.

On the other hand, reflection mode imaging has the potential for wider deployment and utilization. Reflection-mode imaging is compatible with stand-alone imaging and active remote sensing and can be made portable. Reflection-mode imaging requires a radar-like transceiver² and can also be used to map range to targets for 3-D imaging, as well as mapping their velocity. For instance, Cooper et al.² employed a 690 GHz frequency-modulated continuous-wave (FMCW) radar with a ∼30 GHz modulation bandwidth and a 1 m reflector antenna to demonstrate subcentimeter depth resolution and 0.02° angular resolution, forming 3-D images of a target at 25 m. Although not demonstrated, relative velocity can be extracted using the Doppler frequency shift using the same system, employing similar techniques presented using millimeter-wave FMCW radars.^94,95

Fully integrated silicon radar transceivers were employed for reflection mode imaging in the 220–320 GHz band.^19,20,96 A multitransceiver architecture that leverages the large-scale integration capability of CMOS technologies allows for a seamless band coverage from 220 to 320 GHz²⁰ using five 20 GHz bandwidth transceivers, and substrate integrated waveguide (SIW) antennas tuned at five different center frequencies are used to demonstrate a ∼1.5 mm depth resolution.

Moreover, compact pixels integrating a circuit that functions both as a transmitter and a receiver (concurrent transceiver) and an on-chip antenna within an area close to a half wavelength by a half wavelength are demonstrated in the same band.⁹⁷⁻⁹⁹ A fully integrated 430 GHz concurrent transceiver pixel array^21,100 assembled with a Cassegrain reflector antenna²¹ was used for stand-off imaging of a target 3 m away through heavy fog, as shown in Figure 3. With a 6 cm diameter reflector, the measured angular resolution was 0.7°. With a 15 cm diameter reflector, an angular resolution of ∼0.3° that overlaps that of lidars should be possible. The further advances, including the use of the BiCMOS technology that provides SiGe HBTs with a f_max of 720 GHz³⁷ that will enable an increase of output power, will allow an extension of the range beyond 200 m. Such imaging can provide a unique and compelling combination of angular resolution, operation in visually impaired conditions such as fog, smoke, and others, and affordability that no other technologies can match. It should also be possible to increase the operating frequency to the propagation window near 850 GHz to improve the angular resolution by 2×.

A problem with using optics, such as a lens or a reflector, is a limited field of view. A focal plane array of concurrent pixels that can cover a field of view of ∼10° can be used with a Cassegrain reflector without significantly degrading the reflector gain. This, however, is still too limiting for many applications. Mechanical scanning is a straightforward approach to overcome this. However, it increases the system volume, cost, and power consumption, as well as reliability challenges. A way to bypass these is using electronically steerable reflect arrays^101,102 to form external optics. Besides increasing the output power and improving the sensitivity of pixels, realizing affordable and power efficient steerable reflect arrays is a key technical need.

A pair of concurrent transceiver pixels operating at ∼275 GHz fabricated in 65 nm CMOS was also used to form images of targets ∼1 cm away with and without a cardboard cover.¹⁰³ The imaging module, demonstration setup, and images are shown in Figure 4. This suggests that SMMW imagers can be integrated into a form factor consistent with that of smartphones. More in-depth discussions of CMOS SMMW imaging can be found in a paper published by the authors.¹⁰

Figure 4.

275 GHz lensless reflection-mode imaging of targets at 1 cm without and with a cardboard cover. A pair of concurrent transceiver pixels in 65 nm CMOS is employed. (a) PCB and imaging setup, (b) visual image of targets, (c) imaging experiment setups without and with a cardboard cover, (d) image taken without a cover, and (e) image taken with a cover. Reprinted with permission from ref (103). Copyright 2023 IEEE.

Electronic Smelling Using Rotational Spectroscopy

Electronic smelling can be emulated by gas sensing utilizing rotational spectroscopy,⁴ which can be used to identify a wide variety of gases and to quantify their concentrations. Rotational spectroscopy relies on the unique rovibrational absorption fingerprint inherent to each rotating polar molecule. These spectral signatures are known to manifest strong maxima in the SMMW region.¹⁰⁴ Since the absorption lines occupy a very narrow bandwidth (∼1 MHz), frequency-swept continuous wave systems are more suitable than terahertz time-domain spectroscopy systems that rely on wideband pulses with limited repetition rates.¹⁰⁵ Considering how smell is used in daily life, the application opportunities should be almost limitless. Figure 5 shows a typical gas analysis or electronic smelling setup using millimeter wave and SMMW rotational spectroscopy together with a 225–280 GHz receiver¹⁰⁶ and a 205–255 GHz transmitter¹⁸ fabricated in a 65 nm CMOS process. The experimental setup uses a 2 m long gas tube. Figure 6 shows the lines of acetone and acetonitrile located near 240 GHz that have been detected by using the CMOS circuits. The rotational lines are typically around ∼1 MHz wide with a corresponding Q of ∼1000000, and typically, multiple lines are present for given molecules at frequencies between 200 and 500 GHz. Because of these redundancies, rotational spectroscopy can identify a wide variety of molecules in the gas phase, including volatile organic compounds, and quantify their concentrations with almost absolute specificity.⁴

Figure 5.

Rotational spectroscopy setup. (a) Photographs, (b) block diagram, and (c) measured results with and without the breath sample. Reprinted in part with permission from ref (18). Copyright 2016 IEEE.

Figure 6.

Measured spectra using a CMOS transmitter and receiver for acetonitrile (100% absorption) and acetone (1% absorption). Reprinted with permission from Sharma, N.; Zhong, Q.; Choi, W.; Zhang, J.; Chen, Z.; Ahmad, Z.; Medvedev, I.; Lary, D.; Nam, H.-J.; Raskin, P.; Kim, I.; O, K. K. Complementary metal oxide semiconductor integrated circuits for rotational spectroscopy. Proc. SPIE 11390, Next-Generation Spectroscopic Technologies XIII, 113900K, 2020. Copyright 2020 SPIE.

The transmitter RF front-end delivers a saturated output power of −1 dBm. When the input power is −20 dBm, the output power is −8 dBm. The incident power to the sample cell should be less than ∼−20 dBm to avoid saturation of molecules, which limits the dynamic range during concentration measurements. The relatively moderate incident power requirement makes this application particularly well suited for the CMOS realization.

The requirements of a receiver for rotational spectroscopy differ from those for communication, since the receiver needs to detect weak energy absorption dips in the presence of much larger (10⁶ times larger) baseline variations instead of detecting a small signal in the presence of noise. A frequency-modulated (FM) transmitted signal is used to enable detection of small absorption dips. Use of FM also mitigates the impact of the RF output power variations with frequency.^4,107

Typically, rotational spectroscopy systems use atomic clocks as a frequency reference, which is much more costly than that of CMOS transmitters and receivers. This challenge can be overcome by using the fact that rotational lines are narrow and have almost negligible temperature dependence are utilized to demonstrate a frequency reference with long-term stability comparable to that of atomic clocks^18,107,108 and the temperature stability of 3 × 10^–9 between 27 °C and −65 °C, which is approximately within a factor of 10 of that for atomic clocks.^109,110

The performance, including the operation frequency range of rotational spectroscopy IC’s can be further improved by utilizing more advanced circuit and system design techniques. In fact, a multichannel (10 transmitter and 10 receiver) comb-based spectrometer covering 220 to 330 GHz range is demonstrated.¹⁷ Additionally, a dual-band (250 and 500 GHz) spectroscopy transceiver in SiGe BiCMOS is demonstrated.¹⁶

Among many, a particularly exciting application of electronic smelling is analyses of human breath^{16,18,111,112} for disease diagnosis, monitoring of intoxication, environment and fatigue for safety, checking medication compliance for healthcare, and many others. The CMOS transmitter and receiver were also used to analyze human breath.^107,111,113 The measured spectrum for an Ethanol concentration of 38 ppm is shown in Figure 5c. To ascertain the presence of the line, we also measured the spectrum when the gas tube was evacuated. Affordable versatile electronic noses will accelerate the collection of data to improve the efficacy of the technique with the aid of machine learning.²⁸

High Data-Rate Communication

Communication is another key application of SMMWs because of the wide frequency bands that can be allocated. Both high data rate wireless communication^{7,22,23,114−117} and wireline communication^27,118−120 are possible. SMMW wireless links can also be made more energy efficient than 5G links.⁹ An IEEE standard for wireless communication in the frequency range from 275 to 320 GHz is established¹¹⁴ and efforts to adopt 100 to 300 GHz as a part of sixth generation mobile communication standard¹¹⁷ are ongoing.

The greatest challenges of SMMW communication would be limited coverage and high power consumption. They are closely related to the limited performance and efficiency outlined in the second section. Especially when the wide channel bandwidth and/or high-order modulation is employed to support a high data rate, the minimum detectible signal, which is proportional to the product of the bandwidth and minimum required SNR, as in eq 1, increases.

A 240 GHz highly integrated QPSK transceiver for 16 Gbps²² wireless communication and a 300 GHz RF transmitter front-end that can support 105 Gbps (32-QAM) wireless communication²³ have been demonstrated using CMOS. A 110 Gbps QPSK communication at 240 GHz over 1 m is demonstrated using SiGe BiCMOS.¹²¹ Despite the inferior performance and higher power consumption of 315 GHz CMOS transmitters and receivers²⁵ compared to that of 5G 28 GHz CMOS transceivers, the communication capacity efficiency⁹ can be higher than that of 5G transceivers because of the large bandwidth and improved antenna gain with frequency when the antenna aperture size and range are kept constant. Figure 7 shows a schematic and a die-photograph of the 315 GHz CMOS receiver,²⁵ which is fully integrated, synchronizes carrier frequency, and generates a digital output. For an aperture size of 10 cm × 10 cm, an experimentally reported 315 GHz CMOS transceiver can lead to higher energy efficiency up to ∼1000 m compared to that of 5G 28 GHz CMOS phased-array transceivers.⁹ As a matter of fact, the capacity power efficiency⁹ for systems using an active gain with a phased array is multiple orders of magnitude lower than that using a passive gain of aperture such as a Cassegrain reflector. Because of this, low profile, adaptive, steerable, energy efficient, and easy to use apertures are once again a key to implementing high performance SMMW wireless links.

Figure 7.

Fully integrated 315 GHz receiver that synchronizes carrier frequency and generates a digital output. Reprinted with permission from ref (25). Copyright 2020 IEEE.

Wireline communication using a dielectric waveguide with CMOS integrated circuits over a short distance (∼1 m) has been proposed to mitigate the complexity of high data rate communication over copper wires and the challenges for implementing optical communication systems.^{5,26,27,118,122} Frequency division multiplexing and polarization division multiplexing can be simultaneously used to increase the data rate for a given bandwidth.¹¹⁸ Use of horizontal and vertical polarization communication modes in five 45 GHz frequency bands between 157.5 and 382.5 GHz provides a total of 10 channels. With a data rate of 30 Gbps per channel, an aggregate data rate of 300 Gbps could be supported. Figure 8 shows a block diagram of the proposed communication system.¹²³

Figure 8.

Block diagram of the proposed communication system.

A 1 m long and 8 dB loss dielectric waveguide communication system with a bit error rate of 1 × 10^–12 can be realized with a transmitter with an output power of −6 dBm and a demodulator with a DSB noise figure of 18 dB, which have already been demonstrated at ∼300 GHz using 65 nm CMOS.^118,124 Referring back to Figures 1 and 2, 400 GHz transceivers with adequate performance should also be possible.^78,123 The communication range is limited by the waveguide loss of ∼8 dB/m instead of ∼0.2 dB/km for optical fibers. Lowering the waveguide loss is an opportunity for impactful research. Increasing the operating frequency to 1 THz or higher enabled by the advances of high frequency performance of silicon IC technologies will allow uses of waveguides with a smaller cross-section and greater mechanical flexibility as well as uses of waveguide materials that may have lower loss. The discussions in this section were originally published in papers by the authors.^8,9

Conclusions and Future Perspectives

Over the past 15 years, the field of silicon SMMW circuits and systems has seen groundbreaking advances (Figure 9). These were enabled by the availability of high-yielding CMOS and SiGe devices with f_max of ∼350 and 450 GHz, respectively, and interconnect processes that are compatible with SMMW operation. The output power has improved by a factor of ∼1000 to 100000 from −49 dBm at 410 GHz¹²⁵ and −46 dBm at 324 GHz¹²⁶ for MOS transistor circuits and −38 dBm at 590 GHz¹²⁷ and −20 dBm at 280 GHz¹²⁸ for SiGe HBT circuits, reaching −3.9 dBm at 440 GHz for a single unit in CMOS¹²⁹ and 5.2 dBm at 317 GHz for a 16-element array in SiGe BiCMOS.⁷² The sensitivity for a 1 kHz bandwidth signal has improved by ∼100 million times from −49 dBm at 600 GHz (95 dB equivalent noise figure)⁸¹ to −129 dBm at 420 GHz (14 dB noise figure).⁷⁸ Moreover, by employing high-order harmonics, the operation frequency of fundamental circuit blocks such as frequency multipliers and coherent mixers was raised to 1.3⁴⁶ and 1.2 THz,⁴⁷ respectively, while that of amplitude detection was raised to 10 THz^48,80 using a Schottky diode in CMOS. Going forward, as SiGe HBTs with f_T, and f_max of ∼505 and ∼720 GHz,³⁷ respectively, become widely available, the performance of silicon SMMW circuits at a given operating frequency is expected to improve, and the operating frequency at a given performance is expected to improve.¹³⁰ With such transistors, circuits operating well beyond 1 THz with a useful performance will be possible. The state-of-the-art performance is sufficient for many SMMW applications.

Figure 9.

Timeline of the silicon SMMW device and circuits. f_max = maximum oscillation frequency, P_out = output power, and P_sen = sensitivity for 1 kHz signal bandwidth. *Interconnects are de-embedded.

The capabilities that differentiate the silicon SMMW electronics from the others are the integration of complex systems into a small form factor and reduced fabrication cost. This integration not only means the analog and digital signal conditioning and processing subblocks can be integrated to lower the system cost, to further improve yield and performance by built-in correction and to make system implementation easier, but also means that multiple SMMW building blocks can be integrated within a front end with minimal performance penalties associated with the interconnects. Additionally, when the opportunities for broad deployment of affordable SMMW electronic systems emerge, the silicon IC technology infrastructure will be the most capable to support the necessary high-volume and low-cost manufacturing.

Lastly, the choice of particular technology will depend on the applications. For those with moderate performance constraints that require a smaller form factor, low cost, low power, high production volume, or combinations of these, silicon electronics would be the most suitable. This list of examples of such applications, especially accounting for the expected advances of silicon SMMW electronics, may include imaging with a range of up to several hundred meters at 400 GHz and higher, gas sensing up to ∼1 THz, and 100 Gbps+ communication at 300 GHz and higher over a few meters to several kms.

Author Contributions

^§ W.C. and K.K.O. contributed equally to this paper.

This work was supported in part by DARPA, NASA, the Semiconductor Research Corporation (C2S2, TxACE, and ComSenTer), the Naval Research Laboratory, and the Samsung Global Research Outreach and TI Foundational Technology Research Program on Millimeter-Wave and High Frequency Microsystems.

The authors declare no competing financial interest.