A team at the Korea Institute of Science and Technology has published a brain-computer interface chip in Advanced Science that reads both electrical and optical brain signals on a single silicon probe. The paper is titled “Fabrication of High-Density Multimodal Neural Probes Based on Heterogeneously Integrated CMOS,” DOI 10.1002/advs.202524260, in the Wiley journal Advanced Science in 2026. First author is Ju-hee Mun (문주희) of KIST and Korea University. Senior author is Changhyuk Lee (이창혁, also romanised Lee Chang-hyeok in English press), Principal Researcher at KIST’s Brain Science Institute, who leads the research team behind the paper. Cho Il-joo (조일주) of Korea University College of Medicine, who is the Program Director of the K-Moonshot Project’s brain-computer interface mission, is a co-author.
What the chip does
The probe integrates 416 microelectrodes and 832 CMOS light-sensing pixels (photodiode-based) across 13 hair-thin silicon shanks in a single monolithic package. The electrical channel is an intracortical microelectrode array (not ECoG surface arrays, not scalp EEG). The optical channel is CMOS image-sensor-style photodiode readout, designed to capture optical signals emitted from adjacent cells of the kind used with fluorescent activity indicators. The two modalities record simultaneously on the same shank.
This is not fNIRS, not fUS, and not two-photon microscopy. The optical layer sits on the probe itself rather than on external hardware, which is what distinguishes the KIST design from prior multimodal recording setups that combined separate optical and electrical systems.
CMOS heterogeneous integration lets the chip do on-probe signal processing rather than sending raw data off to an external processing unit. KIST claims the fabrication approach reduces cost to approximately one hundredth of existing methods. That figure is KIST’s own institutional claim, not independently benchmarked in the paper.
Preclinical validation
Testing was done in live mice, recording from three brain regions, with one-week biocompatibility follow-up reporting no inflammation and no motor deficit. The paper does not report primate or human data. The regulatory path from mouse validation to human clinical trials for any implantable BCI probe is measured in years, not quarters.
The “world’s first” claim, precisely scoped
KIST claims this as a world-first. The claim is stated in the KIST 5 August 2026 press announcement (verbatim Korean: 세계 최초로 개발했다) and picked up in ChosunBiz English on 5 August and in BigGo Finance’s English rewrite on 7 August. Lee’s own framing describes it as opening a new technology category he calls optoelectronic-semiconductor BCI.
The scope of the “first” claim is narrower than most headline coverage suggests. Multimodal BCI recording has existed for years in various configurations, including hybrid fNIRS-plus-EEG systems and photonic-neural probes that combine light delivery with electrical recording. What is genuinely new here is the CMOS heterogeneous integration of both electrical and optical sensing on a single monolithic probe at this channel density. Prior art comparisons:
μLED-based optoelectronic probes exist for delivering light into brain tissue. They emit light rather than sense it. Implantable nanophotonic neural probes use waveguides for patterned photostimulation with co-recording of electrical activity. Neuropixels (Allen Institute plus IMEC) is silicon plus CMOS plus high-density electrodes, with no optical channel. KIST’s novelty is on-probe optical sensing at high density using CMOS photodiodes.
BigGo Finance’s headline framing that “Elon Musk couldn’t crack neuron classification” is editorial spin, not a KIST claim. Neuralink and other electrical-only BCI operators do spike-sort (temporal firing pattern discrimination) as a matter of course. What their electrical-only recording cannot do is cell-type discrimination via fluorescence readout. That is the gap KIST’s design is aimed at, and it is a genuine gap, but the framing needs to be precise: cell-type discrimination through optical signature, not the broader claim that Neuralink cannot classify neurons.
Where KIST sits in Korea’s BCI ecosystem
Cho Il-joo’s co-authorship is the institutional bridge. He runs the K-Moonshot Project’s BCI mission, which launched on 11 March 2026 as one of twelve moonshot missions coordinated by the Ministry of Science and ICT, with total budget around 600 billion won across the twelve missions from 2026 to 2030. The budget is not BCI-specific, but the BCI mission is one of the twelve funded lines. Cho’s public timeline commits to invasive-BCI commercialisation for quadriplegic use within two to three years, and general brain disease within five.
Ybrain, the tDCS device maker whose MINDD STIM tDCS device targets major depressive disorder, is the K-Moonshot BCI mission’s sole industry-side partner per the 11 March 2026 memorandum of understanding, and is one of fourteen founding entities in the Korea BCI Industry Association that held its inaugural meeting on 16 June 2026 in the Centerfield building, Gangnam-gu, Seoul. The Association is chaired by Baek Nam-jong, President of Seoul National University Hospital and rehabilitation medicine professor. Kim Yong-jin, chair of ISO/IEC JTC 1 SC 41 (Internet of Things and Digital Twin), is the standing vice president. Founding entities span industry (Ybrain, Dynamic Solution the former Neopect, Ceragem among others), medical (Seoul National University Hospital, Severance, Korea University Anam Hospital), and academic (KAIST, DGIST, UNIST, Yonsei, Hanyang, Korea University and Catholic Kwandong).
Where Korea sits in the Industrial Builder cohort
The Issue 01 four-jurisdiction analysis grouped Korea with China as Industrial Builder. That grouping is correct on posture and institutional scaffolding: K-Moonshot mission structure, Association-led standards work, chaebol-adjacent capital access, and named clinical-medical bridges through SNUH and Severance. It is not correct on invasive-BCI commercial trajectory.
China has Neuracle NEO NMPA-approved on 13 March 2026, with the first commercial prescription written at Huashan Hospital on 13 July 2026, National Healthcare Security Administration pricing guidance in April 2026, and Neuracle’s STAR Market IPO prospectus accepted 11 June 2026. Korea has an approved electroceutical (Ybrain’s MINDD STIM tDCS, MFDS Class 3 approval April 2021, deployed at 767 hospitals and clinics) and institutional scaffolding through K-Moonshot plus the Association, but no commercial invasive-BCI approval from MFDS yet.
Cho Il-joo’s own two-to-three year invasive-commercialisation timeline for quadriplegic use dates the earliest realistic commercial-invasive-BCI approval in Korea to 2028 or 2029.
What to watch
Three near-term milestones matter for reading the KIST publication properly.
First, whether Lee’s team advances the chip into non-human primate work. Mouse-to-primate is the next translational gate. Timeline for BCI chips of this class from mouse validation to first primate implant is typically twelve to twenty-four months, depending on regulatory pre-clearance requirements and available NHP surgical infrastructure.
Second, whether KIST or Korea University spins out a company around the optoelectronic-semiconductor BCI category Lee is naming. K-Moonshot mission funding suggests spinout capital is available. A named spinout with named commercial partners would move this from research announcement to operator signal.
Third, whether other academic teams globally replicate the CMOS multimodal integration approach. Allen Institute plus IMEC (Neuropixels), Neuralink internal engineering, Paradromics’s Connexus platform and various academic photonic-neural probe groups are the plausible fast-followers. Twelve months from publication is a reasonable window to expect the first citing follow-on paper.
This is a research milestone. Commercial translation is what to watch next. What the paper demonstrates is that Korea’s academic-industrial BCI base is producing frontier hardware capable of extending the state of the art beyond the electrical-only paradigm that dominates the current commercial cohort. Whether that becomes an operator-tier capability in Korea, or migrates to a US or Chinese commercial implementation first, is the open question the paper leaves on the table.