In 2013, Safaie et al. proposed an integrated EEG–fNIRS system [38]. This system consisted of four primary parts: front-end EEG recording electronics, head-mounted optoelectronics, a control unit, and a laptop. In the fNIRS block, there were eight dual-wavelength LEDs (760 nm and 850 nm, L760/850, Epitex) as sources, and four silicon diodes (ODA-6WB-500M, Optodiode) employed as detectors. This arrangement could produce up to 32 channels with a theoretical source-detector separation (SDS) in the range of 20–63 mm. A 16-bit analogue-to-digital converter (ADC) (ADS1178, Texas Instruments) was utilised for fNIRS data acquisition, and the overall sampling rate was 8 Hz. The EEG block supported 12 standard EEG Ag/AgCl sintered cup electrodes. A 16-channel biopotential front-end amplifier was chosen to acquire EEG signals which could measure electrode–skin impedance in the range of 0–100 kΩ. A 16-bit ADC (AD7985, Analog Devices) was multiplexed and utilised to acquire the recorded EEG signals. The sampling rate of EEG signals was 1024 Hz. A custom-designed optoelectronic “patch” was implemented to integrate both EEG and fNIRS components together, with dimensions of 35 × 80 × 10 mm³ and weight of 90 g. The control unit and laptop were utilised for data acquisition and transmission. This system can be wirelessly operated via an in-built Bluetooth module, demonstrating a total system power consumption of ~400 mW.

This system successfully integrated EEG and fNIRS modules into a wearable system. It achieved wireless operation and a relatively lightweight form factor. However, there are also several concerns with this work. First, the fNIRS module claimed a promising SDS (up to 63 mm) and a theoretical dynamic range (up to 198 dB), yet no demonstrations were conducted at longer separations to validate the data quality and practical dynamic range. Besides, the capacity for (3D) imaging of the fNIRS module was unclear. Moreover, the EEG demonstrated a relatively limited measuring range of electrode–skin impedance, which is inadequate for use of dry electrodes. Besides this, the EEG module and fNIRS module separately employed two different ADCs for data acquisition. This design not only additionally increased the size and power of the system, but also could cause concern for timing precision (between markers in the fNIRS and EEG data streams). More critically, for both fNIRS and EEG modules, this system applied extensive cabling for analogue signal transmission from EEG electrodes and optical detectors to the distant control module, which could be inconvenient for practical applications as well as make the system vulnerable to radio frequency (RF) noise. In addition, the scalability of current ergonomic design of the system is limited.

The same year, building upon their prior work [41], Sawan et al. developed a wireless, wearable EEG–fNIRS integrated system [39], as shown in Figure 1a below. This system contained two primary parts: a wearable helmet and a control unit. In the helmet, there were eight sources, eight detectors, and eight EEG recording sites. Dual-wavelength LEDs (735 nm and 850 nm, Epitex) were chosen as sources, and avalanche photodiodes (APDs) (S2384, Hmamatsu) were selected as detectors. Particularly, a high bias voltage (150 V) was applied to the APDs so as to achieve satisfactory sensitivity. Theoretically, each light source could be coupled with four surrounding detectors, thus this system could produce up to 32 fNIRS channels, but the actual value of SDS was not noted. The front-end LED driving circuits and detection circuits were both fitted into small-size circular PCBs (130 mm²). However, the physical implementation of the EEG electrodes and their front-end electronics were unclear. In the control unit, three different stack layers of PCBs were implemented to realize system control, power, and data transmission. A 16-bit ADC was utilized to acquire both fNIRS and EEG data, with sampling rates of 20 Hz and 320 Hz, respectively. A Bluetooth module was embedded into the control unit to achieve wireless data transmission. The control unit contained a Li-ion battery (10Ah-7.4 V, 76 × 80 × 40 mm², 400 g) to act as supply power for the overall system, with a measured power consumption of 2.2 W. The overall dimensions of the control unit were 160 × 130 × 82 mm², and the weight was 800 g. This system has recently been further upgraded to include more sources, detectors, and electrodes, achieving 128 fNIRS channels (32 sources, 32 detectors) and 32 EEG channels, but with increased size and weight, along with a more extensive cabling [42], as shown in Figure 1b.

This work demonstrated a wireless, wearable, integrated EEG–fNIRS system. A shared ADC architecture was implemented to simultaneously acquire fNIRS and EEG data, and this setting could potentially improve the timing precision of multimodal data acquisition. However, similar to Safaie et al.’s work presented above, this system employed highly cumbersome cable connections between front-end components and the control unit. This setting would limit its practical and longer-term applications, and, more critically, could incur external RF noise during the analogue data transmission between the helmet and the control unit, and then consequently bias the data quality. This work explained the implementation of head-mounted optoelectronics, but the realization of EEG electrodes and associated front-end electronics was not clearly demonstrated. Moreover, the measuring range of electrode–skin impedance was not presented either, so no clear information about the compatibility with different types of electrodes (i.e., wet, dry, or both) could be provided for users/readers. Besides this, the SDS value of fNIRS data was also unclear, and the possible imaging capability was also not demonstrated. In addition, the crosstalk effect between optics and electronics of this system was not evaluated.

In 2017, von Luhmann et al. developed a modular system architecture to achieve an integrated multichannel EEG–fNIRS system [40], as shown in Figure 1c above. Each module was equipped with two dual-wavelength LEDs (750 nm and 850 nm, L750/850-04A, Epitex) as NIR light sources, and employed two silicon photodiodes (OPT101, Burr-Brown) as NIR detectors. The SDS was arranged around 35 mm. Each module could produce up to six fNIRS channels (four inter-module channels, and two theoretical cross-module channels). A 24-bit ADC (ADS 1299, Texas Instruments) was utilised for fNIRS data acquisition. The sampling rate of the fNIRS data was set as 16.6 Hz. The noise equivalent power (NEP) of the detection circuitry was characterised, obtaining minimum value of 4.77 pW at 850 nm and maximum value of 5.92 pW at 750 nm. In each module, up to six theoretical EEG channels could be generated. The 24-bit ADC used for fNIRS data acquisition was also utilised for EEG data acquisition simultaneously, achieving a sampling rate of 500 Hz. Due to the high input impedance (1 TΩ) and common-mode rejection ratio (–110 dB), both wet and dry EEG electrodes could be (theoretically) suitable for use. A three-axis accelerometer (ADXL343, Analog Devices) was also embedded into each module for local movement monitoring. Besides, a Bluetooth module was implemented for wireless data transmission. The module can be powered by a small-size Li-ion battery (300 mAh, 28 × 34 × 0.6 cm³), and the power consumption of each module was around 360 mW. This modular architecture can be theoretically scaled up to four modules, and in this presented work, a three-module fNIRS+EEG system was demonstrated (as shown in Figure 1c), producing 13 fNIRS and 8 EEG channels.

This work proposed a modular, wearable, multimodal system that can achieve fNIRS measurement, EEG recording, and motion monitoring concurrently in a single module. Besides this, they created a shared ADC architecture which could potentially improve time precision and data synchronisation. In addition, battery powering and wireless operation were also implemented for each module. This system demonstrated several merits, however, there are still some key limitations. First, the capability for (3D) brain imaging was yet demonstrated. Particularly, (most of) the SDS were fixed at 35 mm, which indicates that only sparse spatial sampling could be produced. Moreover, the footprint of each module was large (42 × 42 mm²), which constrained the wearability and comfort of the system. Furthermore, the scalability of this modular design was limited, and it is challenging to use this system for whole-scalp sampling. In addition, the weight of the module was unclear.

In 2019, Lee et al. proposed a dry electrode-based integrated EEG–fNIRS system [43]. A block diagram of the system is shown in Figure 1d. In this system, two dual-wavelength LEDs (730 nm and 850 nm, OE-MV7385-P, Opto ENG) were utilised as fNIRS sources, and six silicon photodiodes (OPT101, Texas Instruments) were employed as fNIRS detectors. This arrangement generated eight fNIRS channels in the area of 9 cm × 3 cm, with a fixed SDS of 27 mm. An MOSFET-based LED driver was implemented to modulate LED illuminations with fine-tuned intensity. A 16-bit ADC (ADS8688A, Texas Instruments) was embedded to acquire the eight-channel fNIRS data with a sampling rate of 5 Hz. A custom-designed prototype of dry electrodes was implemented for EEG recording. Each dry electrode consisted of 18 spring-loaded probes (SK100R, Leeno Industrial Inc.) so as to provide contact with the subject’s scalp. Sixteen dry electrodes were employed in this system for EEG recording. Two 24-bit ADCs (ADS1299, Texas Instruments) were dedicated to being used for 16-channel EEG data recording, with a sampling rate of 250 Hz. Two four-layer PCB-based control units (70 mm × 70 mm each) were fabricated for data acquisition for all the EEG and fNIRS channels and overall system control. This system can be battery powered, and the average power consumption of each recording channel was about 18.8 mW.

Although this work successfully integrated fNIRS and EEG modalities into a standalone system, this system has several limitations similar to the systems of [48,49,50] (Table 2). First, it utilised extensive cabling between optodes/electrodes and the distant control unit, which was bulky and cumbersome, and limited its subject comfort and long-term wearability. More critically, this could potentially make both fNIRS and EEG signals vulnerable to environmental RF noise. Moreover, this system demonstrated a fixed SDS of only 27 mm, thus, only sparse sampling can be achieved using this device. In addition, the total weight of the system was unclear. Overall, this newly developed system did not demonstrate obvious advancements over previous designs in [38,39,40,41,42].

Of note, there are several commercially available wearable fNIRS systems [48,49,50] that claimed that they can be integrated with EEG, yet these systems only mechanically integrated fNRIS optodes with EEG ports on the cap. These systems still employed separated electronic designs, far from a fully integrated system.

The section above described the integrated EEG–fNIRS technologies, which fully consisted of off-the-shelf discrete components. An alternative approach is to implement and integrate the primary fNIRS measurement circuitry and EEG recording circuitry into a single microchip using custom-designed CMOS integrated circuits (IC). This strategy can potentially allow for a smaller footprint, ultralight weight scale, lower noise and power, better signal quality, and improved timing precision.

In 2011, Chua et al. proposed a CMOS-based, highly integrated EEG–fNIRS system [44]. In this system, six dual-wavelength LEDs (735 nm and 890 nm) were equipped as fNIRS sources, and twelve detectors were positioned adjacent to these LEDs, as shown in Figure 2a below. The fNIRS sensor array was implemented on a bendable PCB. The SDS was set as 14.14 mm, and this setting generated 24 source-detector channels. A 10-bit ADC was implemented to achieve a 1 Hz sampling rate for fNIRS data. Another 10-bit ADC was employed to record EEG data at a sampling rate of 128 Hz. However, the number of EEG channels and electrodes were not depicted. A UMC 65 nm CMOS technology was used to fabricate the microchip that included an fNIRS data processor, a four-channel independent component analysis (ICA) (for EEG recording), and a heart rate variability (HRV) analysis processor. The dimensions of the die were 1317 um × 1317 um, and the active area (core size) was 680 um × 680 um. The simulated power consumption of the microchip was 3.6 mW.

This work presented a microchip-based multimodal platform for simultaneous fNIRS and EEG measurement and data acquisition. Particularly, it integrated an fNIRS data processor, an ICA engine, and a HRV processor into a compact chip layout. However, there are several concerns with this work. First, this work did not clearly demonstrate how the fabricated chip can be employed as a practical multimodal system. Secondly, the implementation of EEG front-end and electrodes was unclear, and the number of EEG channels was also not provided. Moreover, the possible crosstalk between fNIRS signals and EEG signals was not taken into consideration. Furthermore, this work demonstrated a bendable PCB-based sensor array, but it was unclear how this array can be physically and practically applied onto patients/subjects. Besides this, the SDS of the array was fixed at 14.14 mm, and this indicates that it would be challenging to apply this system to produce (3D) images. In addition, all the data presented in this work were based on simulation, while measurement validations of the system performance were not demonstrated.

In 2016, Ha et al. proposed an integrated EEG–fNIRS ear-module system [45], as shown in Figure 2b. A system-on-chip (SoC) architecture was implemented, containing three primary blocks: (1) a sophisticated front-end transimpedance amplifier, (2) a tuneable VCSEL driver, and 3) a reconfigurable impedance boosting loop for EEG recording. The SoC was then integrated into a wearable ear-module device, which primarily included an ear hook and an earpiece. In this ear-module device, there was one dual-wavelength VCSEL (670 nm and 850 nm) utilised as optical source and two photodiodes employed as detectors. In particular, one of the photodiodes was closely placed adjacent to the VCSEL (located at an optode board) to act as a dummy detector while another photodiode located at the ear hook served as the main detector. An 11-bit ADC was realized and shared to record both fNIRS and EEG data. A dedicated VCSEL driver was implemented with optimized SNR, and a 7-bit digital-to-analogue converter (DAC) was incorporated with the VCSEL to modulate the light intensity. For EEG recording, a reconfigurable impedance boosting loop was proposed that can potentially achieve input impedance (of EEG) up to 5.4 GΩ. This ear-module SoC was fabricated in a 110 nm 1-poly 6-metal CMOS process. The die size area was 450 um × 2250 um. The maximum power consumption of this ear-module SoC was 46.2 mW.

Although this system demonstrated some interesting features, such as the SNR-optimizing tuneable VCSEL driver for NIR light illumination and the reconfigurable impedance boosting loop for ultra-high EEG input impedance, there are some concerns with this work. First, it was unclear whether this ear-module design can be scaled up to multiple modules for brain monitoring. Secondly, some key parameters of fNIRS were not provided, such as the SDS and the dynamic range, thus it is challenging to evaluate the performance of fNIRS functionalities. Besides this, the sampling rates of fNIRS and EEG measurements were not noted. Moreover, the power consumption of this ear module was relatively high, and the total power consumption could be a concern if the system was scaled up to include more modules. In addition, the weight of this system was unclear.

One year later, the same group in [45] above proposed a different integrated EEG–fNIRS system, using more advanced 65 nm CMOS technology [46]. Figure 2c shows the EEG–fNIRS integrated system for anaesthesia depth monitoring. In the proposed monitoring system, a dual-wavelength VCSEL (670 nm and 850 nm) was utilised as a source, and a photodiode was employed as an NIR detector. There were four electrodes (two recording electrodes, one ground electrode, and one reference electrode) to obtain two EEG channels. A polyethylene terephthalate (PET) film and a flexible PCB board were fabricated to house the fabricated SoC, NIRS sources and detectors, EEG electrodes, and accessory components, achieving a lightweight form factor (<26 g) and a relatively small profile (35 mm × 260 mm) of this monitoring system. The core SoC, containing the main circuitry for EEG–fNIRS recording, data acquisition, and system control, was embedded on the flexible PCB section of the wearable monitoring system, as shown in Figure 2c. The proposed SoC contained five primary functional blocks: (1) EEG recording module; (2) NIR detection module; (3) VCSEL driving circuitry; (4) ADC; and (5) digital control/communication module. In the EEG module, a reconfigurable impedance boosting loop was adapted from [45] to achieve a 1 GΩ input impedance. A logarithmic transimpedance amplifier (TIA) was implemented in the NIR detection module so as to obtain a dynamic range of 60 dB. A 6-bit DAC was employed in the VCSEL driving circuitry to modulate the drive current from 0 to 17.2 mA to modulate the light intensity. A shared 12-bit ADC was utilised to simultaneously fetch both fNIRS and EEG data, with a sampling rate of 20–80 Hz and 2 kHz, respectively. A digital module was realized in the SoC to achieve communication, buffering, filtering, and system control. This SoC was fabricated in a 65 nm CMOS process, with a compact layout of 4000 um × 4000 um. A Bluetooth module was embedded in the system to support wireless operation. The maximum power consumption of the system was 25.2 mW.

This work demonstrated a sophisticated SoC for integrated EEG–fNIRS monitoring. It realized a lightweight, wearable system for accurate anaesthesia depth monitoring. Though promising, this work has several limitations similar to the work in [45] above, mainly including (1) uncertainty of the scalability for large-area, or even whole scalp, monitoring; (2) unclear information for SDS; (3) relatively restricted dynamic range (60 dB) for NIR light detection that can limit wider applications of this system.

In 2018, Xu et al. proposed a CMOS-based EEG–fNIRS–electrical impedance tomography (EIT) integrated system [47]. fNIRS, EEG, and EIT functional blocks were implemented and integrated into a microchip. In the fNIRS functional block, a dual-wavelength LED (735 nm and 850 nm) was utilised as an NIR source, and a silicon photomultiplier (SiPM) (with 4871 square microcells, 5.1 mm × 5.1 mm) was employed as an NIR detector, with a bias voltage of 30 V and a power consumption of ~1.5 mW. A 12-bit ADC was utilised to fetch the fNIRS data. Figure 2d shows an fNIRS patch with sources and detectors. In this fNIRS patch, there were two source locations and two detector locations, with a fixed SDS of 30 mm. For EEG recording, it achieved an input impedance of 720 MΩ, but no practical implementation of EEG electrodes was demonstrated in this work. The presented microchip was fabricated using a 180 nm CMOS technology, with a compact chip layout, a small die size (4000 um × 4000 um), and a low power consumption of 665 uW (chip only).

This presented work achieved an integrated design for multimodal EEG–fNIRS–EIT monitoring and demonstrated an advanced microchip design. Despite its merits, there are still some concerns with this work. Firstly, this microchip design can only support one fNIRS channel and one EEG channel; more chips would be needed with more measurement channels. From the primary fNIRS patch design shown in Figure 2d, the scalability of this design is unclear, as well as how one could extend this microchip to a practical, modular, wearable EEG–fNIRS system. Besides this, the fixed SDS setting could limit the sampling capability of the system. Long-term wearability and subject comfort could be additional concerns. Moreover, although the utilisation of SiPM could potentially improve the NIR detection sensitivity, the required relatively high bias voltage (of 30 V) could be a concern when mounting the optodes on subjects’ heads. In addition, the possible crosstalk between fNIRS and EEG was not properly taken into account.