Over the past two decades, the MicroLED sector has repeatedly told the same story: smaller, brighter, and more power-efficient, only to get stuck on yield. Attributing the problem entirely to "making displays the way light bulbs are made" is incomplete. The process inertia of lighting LEDs has indeed hindered display applications, but the yield hell of MicroLED is shaped by at least several bottlenecks: the cost structure resulting from small-size substrates, the precision and speed of mass transfer, red light efficiency, driving and detection repair, and uniformity at the million-pixel level. Size is a critical lifeline among them, but not the only answer.
StarkSEMI has chosen to enter the market with 8-inch silicon-based gallium nitride (GaN).
Founded in October 2024, according to public information, in September 2025, the company built China's first 8-inch silicon-based GaN MicroLED pilot production line covering the entire process of epitaxy, heterogeneous integration, and chip testing; in May 2026, it released a full series of micro-displays in red, green, and blue primary colors at SID Display Week, and exhibited the "dual red light" technology route on the same stage; in September 2026, it completed a financing of over CNY 1 billion, led by CATL Puquan Capital, with Meituan Longzhu, Guotai Haitong, Youshan Capital, CDH Hong Kong, and state-owned capital from Chongqing, Jiangsu, and other places joining in. Existing shareholders such as Hillhouse Ventures, Sequoia China, and Matrix Partners China increased their investments, and the cumulative financing amount has exceeded CNY 2.5 billion.
But financing is not the beginning of the story. The real variable is its attempt to drag MicroLED out of the small-size process inertia of lighting LEDs and into the 8-inch manufacturing system of silicon-based semiconductors.
| Size Is a Critical Lifeline, But Not the Only Answer
Wafer size almost defines the cost structure of a process. The effective area of an 8-inch wafer is about 1.78 times that of a 6-inch wafer and 4 times that of a 4-inch wafer. The larger the wafer, the smaller the proportion of edge waste, and the lower the cost allocated per unit chip. This is common sense in the silicon industry.
Traditional MicroLED is largely trapped at the 4-6 inch scale for three reasons: sapphire substrates are insulating, brittle, and have poor thermal conductivity, making it difficult to grow large-size uniform wafers; supporting MOCVD, lithography, and bonding equipment are mostly developed for small sizes; and the micron-level precision, million-pixel uniformity, and PPM-level defect control required for display-level applications have a narrow process window at small sizes. The combination of these three factors makes the industry prone to getting stuck in the valley of death of small-scale verification.
The significance of 8 inches lies in its impact on the threshold of consumer-grade AR: near-eye displays require extremely high pixel density and optical efficiency, and the effective area of 4-6 inches is insufficient to amortize production line costs, let alone achieve consumer-grade pricing. With each step up in size, the unit cost curve shifts downward.
However, two things must be made clear. First, an 8-inch scale does not automatically solve mass transfer. Transferring millions of micron-scale LED chips onto the driving substrate typically requires a yield of over 99.99%, which has no direct causal relationship with wafer size. Second, the 8-inch route itself is not without controversy. JBD chose to skip 8 inches and directly pursue 12-inch wafer reconstruction, citing publicly that there is a size mismatch between the epitaxial wafer and the silicon-based backplane, leaving about 56% of the silicon-based backplane area unutilized, and that the yield of 8-inch wafers in key processes such as bonding and substrate lift-off is relatively low. StarkSEMI chose 8 inches, JBD chose 12-inch reconstruction, and Porotech partnered with Foxconn to build the industry's first 8-inch platform specifically designed for MicroLED, reducing the pixel size to 1.25μm. The size route is still diverging; 8 inches is an option for infrastructure upgrade, not the only solution.
| Why Silicon-Based GaN
StarkSEMI has chosen a more unconventional path: growing gallium nitride (GaN) on silicon substrates. Silicon has been the universal language of the semiconductor industry for decades; equipment for crystal growth, polishing, epitaxy, lithography, and bonding for 8-inch silicon wafers is all readily available, and the yield system has been validated across countless generations of devices. Growing GaN on silicon is equivalent to borrowing the muscle memory of the entire silicon industry—precisely what MicroLED lacks most: a manufacturing paradigm that can be scaled and replicated. Innoscience, previously founded by StarkSEMI's founder Luo Weiwei, has become the world's first company to achieve mass production of 8-inch silicon-based GaN wafers, and the feasibility of this route has already been proven in the power device sector.
The challenges are also straightforward: the lattice and thermal expansion coefficients of GaN and Si are mismatched, making the epitaxial layer prone to cracking and warping. This is precisely the value of a full IDM process—StarkSEMI has built the entire chain from epitaxy and chip manufacturing to wafer bonding, keeping these three processes in-house to lock in uniformity when scaling up the size; otherwise, "8 inches" would just be a gimmick. The pilot line has a planned annual capacity of approximately 12,000 wafers, and the production line is highly compatible with 8-inch CMOS semiconductor chip processes, meaning it can leverage the accumulated experience of mature CMOS production lines for equipment adaptation and process stability.
Red light is the toughest challenge in this equation. The bottleneck for full-color display has long been red light efficiency: the external quantum efficiency (EQE) of AlGaInP can exceed 60% at large sizes, but when miniaturized to the micron scale, the proportion of sidewall non-radiative recombination increases, causing a sharp drop in efficiency—when the size is reduced to below 5μm, the EQE typically falls below 1%. InGaN red light has the potential to maintain good efficiency under miniaturized conditions, but the current peak EQE remains below 10%. StarkSEMI is simultaneously betting on two routes: Si-substrate ternary InGaN red light focuses on chip miniaturization, process compatibility, and large-scale cost control, while GaAs-substrate quaternary AlGaInP red light focuses on light-emitting external quantum efficiency, wavelength accuracy, and color saturation. Both routes come with their own trade-offs, and neither is fully mature. More critically, there is the integration challenge: AlGaInP cannot be grown directly on silicon-based GaN, requiring hybrid integration or bonding processes to combine with the 8-inch silicon-based platform. The engineering complexity of this step cannot be underestimated. The dual-route bet is logically sound, but there is still extensive engineering validation required between "having a route" and "achieving integration."
| Production Line Precedes Financing, But the Production Line Is Not the Endgame
In September 2025, the 8-inch pilot line was commissioned. A pilot line is not just a PowerPoint presentation; it is a real production line capable of stably running through the entire process. According to public information, this production line covers key stages such as GaN epitaxial growth, chip manufacturing, and wafer bonding. Monochrome MicroLED chips have entered small-batch production, and dual-color and tri-color samples have begun to be sent to leading customers for verification. It must be clarified that while the pilot line covers epitaxy, chip manufacturing, and wafer bonding, public information does not specify whether mass transfer and detection repair are fully integrated into the line—these two stages typically involve independent equipment and process systems, with specialized equipment vendors in the industry providing solutions.
This is the confidence behind the over CNY 1 billion financing.
The logic behind CATL Puquan leading the investment is clear: AR near-eye displays, automotive HUDs, and AI computing optical communications are all potential trillion-dollar scenarios requiring high-density, low-power optoelectronic chips, and 8-inch silicon-based suppliers mastering the full IDM process remain scarce in China. The collective follow-on investment by existing shareholders such as Hillhouse and Sequoia is more convincing than the entry of new money—it demonstrates that the previous round invested not in a story, but in tangible progress witnessed firsthand.
The successful commissioning of the pilot line proves that "the process is feasible," which does not equate to "cost-effectiveness achieved," "yield targets met," or "large-scale mass production resolved." StarkSEMI founder Luo Weiwei stated that this round of financing will focus on capacity expansion, chip yield optimization, and tackling key technologies for full-color displays—this itself indicates that yield and full-colorization remain ongoing processes requiring continuous investment, rather than completed milestones.
| The Second Table Beyond Displays
MicroLED optical communication and micro-display technologies share the same origins, both relying on high-density micron-scale light-emitting arrays and silicon-based integration. Streamlining the display process on an 8-inch platform indeed holds the potential for horizontal migration into optical interconnects. The optical communication between chips and between racks, driven by the explosion in AI computing power, is precisely where micro-array light-emitting devices can shine. StarkSEMI has already sent samples to industry partners to conduct verification testing for optical communication products.
However, "shared origins" does not mean it "can be done as a byproduct." The requirements for optical communication devices regarding modulation rate, coupling efficiency, and long-term reliability differ vastly from those of display panels, necessitating independent device design and verification processes. A more accurate statement is: the 8-inch platform provides horizontally expandable platform capabilities, but whether the second table can be fully occupied still depends on independent technological breakthroughs.
| Size Is the Entry Point
Returning to the beginning—making screens the way light bulbs are made is the wrong approach. But the error lies not just in the method, nor merely in the unit of measurement. The real issue is this: lighting LEDs pursue small sizes, low precision, and cost orientation; display MicroLED requires a large-scale manufacturing system, micron-level precision, million-pixel uniformity, and reproducible yields. 8-inch silicon-based GaN shifts the scale from the 4-6 inches of the lighting industry to the 8-inch silicon of the semiconductor industry—this is the entry point for a paradigm shift.
But an entry point is not the finish line. Scaling up the size addresses the cost structure and equipment ecosystem, while mass transfer, red light efficiency and integration, driving compensation, and detection repair equally determine the ultimate winner. StarkSEMI's value lies in not just telling a story about a smaller light bulb, but attempting to replace the entire production line with a semiconductor-grade foundation. While the sector is still struggling with 6-inch yields, 8 inches provides a new infrastructure. Infrastructure does not equal victory; it merely makes victory more probable. Along the industry's various routes, 8-inch, 12-inch reconstruction, and hybrid integration each have their own logic and stakes—StarkSEMI represents one characteristic route, not the only answer.