2.1.

Multi-beam Mask Writing, the Fundamental Concept

Although many different designs for multi-beam writing were developed in the early 2000s,⁴ the design approach of IMS Nanofabrication was the only successful one; it has been widely adopted by the industry since 2016.⁵ In 2022, NuFlare Technology (NFT) also released a multi-beam mask writer to the industry. This system is based on the same fundamental IMS concept, illustrated in Fig. 1.

Fig. 1

Fundamental concept of multi-beam mask writing.

A single electron source generates a single beam of electrons. This beam is then collimated to a wide parallel beam in the condenser lens system. This wide parallel beam passes through the aperture plate system (APS), which consists of an aperture plate (APP) and a blanking chip (BLC).

The APP divides the beam into more than 260 thousand individual beamlets that then pass through the BLC. Using CMOS electronics combined with micro electro mechanical system (MEMS)-formed electrodes in the BLC, each individual beamlet can be deflected to hit the beam stopping plate in the lower section of the column. This allows for effectively switching individual beams on or off because, if deflected, they do not reach the substrate. By controlling the time of each individual beam’s deflection, different gray-levels, e.g., dose-control, can be realized.

Following the beam path through the projection optics, the system performs a 200× demagnification of the beam array onto the substrate. In tandem, the XY-Stage moves the reticle during exposure, and the beam steering deflector positions the beam array to precisely raster the designated exposure area. This approach effectively realizes a massively parallelized rastering spot-beam writer.

2.2.

Different Approaches to Acceleration Optics and Implications

Both systems available today are using the same design approach, as described in Sec. 2. Furthermore, the detailed configurations of the most relevant aspects are fully identical. This includes the acceleration voltage, demagnification factor, aperture sizes, beam pitch, number of beams, size of beam array, and consequently the size of the image field at the substrate. However, despite the many similarities, there are some significant differences that are crucial for the performance of the writer and even more for the scalability of the technology to future requirements. These differences are outlined in Fig. 2.

Fig. 2

Two concepts of the optical system and beam acceleration method used in today’s MBMWs. Concept A is the simpler design, whereas concept B is more complex and has clear advantages realized by a two-stage acceleration system and a stack of magnetic and electrostatic immersion lenses.

Fig. 3

Three important components of blur in multi-beam mask writing: (a) spherical aberration blur, (b) distortion-related positional blur, and (c) space charge-induced blur by beam current I.

2.3.

Superior Lifetime of Critical Components due to a Two-Stage Acceleration System

Although concept A, followed by NFT allows for a simple design in which the beam is fully accelerated to 50 keV above the APS, the design invented by IMS Nanofabrication, concept B, utilizes a two-stage acceleration in which the beam is initially accelerated to 5 keV and the acceleration to full energy is carried out within the projection optics.

At first glance, the simpler approach of concept A seems appealing and beneficial due its simplicity, as claimed recently.⁶ However, concept B has significant advantages with regards to the longevity of APS lifetime and general optics performance.

The first benefit is the absence of x-ray radiation at the APS in concept B. Here, the energy level remains adequately low; consequently, there is no ionizing radiation created at this stage that could harm the sensitive CMOS electronics of the BLC.

This allows concept B to perform without any risk of electronics damage for many years. The fleet-wide average lifetime is above 2.5 years; following a predictive maintenance approach, individual units have exceeded 5.5 years without any issues.

Although earlier tools following concept A could not deal with the radiation damage of the BLC electronics, recently lifetimes of 1 year were reported using mitigation strategies that limit the impact by radiation.⁶ Reducing the amount of ionizing radiation seems to be managed, but any remaining radiation can create electronics damage, leading to a slow but continuous APS degradation over time and thus impacting the write-performance quality such as the critical dimension uniformity (CDU) and pattern fidelity.

Finally, the more robust design following concept B allows for longer APS lifetimes and results in lower maintenance needs and potential cost savings by extending the components’ productive lifespan.

In addition to the APS lifetime, there is a second advantage in concept B that lies in the control of the beam acceleration within the projection system using a strategically placed combination of electrostatic immersion lenses and magnetic lenses, enabling reduced aberrations and beam distortion-related blur and thus superior imaging quality.

2.4.

Superior Aberration and Distortion-Related Blur due to Stack of Magnetic and Electrostatic Immersion Lenses

Different from light optics, in charged particle systems the resolution is not dominantly dependent on the illuminating wavelength. In charged particle systems, the Coulomb blur becomes the dominant factor to limiting the resolution when increasing the beam intensity required for high-throughput productivity. Accordingly, all other contributions to blur need to be reduced to the very minimum to achieve the minimum half pitch and assist features.

The following three contributors to blur (Fig. 3) define the resolution performance of multi-beam technology: spherical aberration blur, distortion-related positional blur, and Coulomb interaction or space charge blur. Each of these contributors needs to be minimized in best performance MBMW.

The two-stage acceleration system allows for the utilization of combinations of electrostatic immersion and magnetic lenses and presents distinct advantages in managing these blur types. It enables effective aberration correction and allows for the minimization of both aberration blur and beam field distortion at the same time.

Spherical aberrations, meaning a different focal length for paths close to electron-optical axis compared with paths further away from the axis, are a widely known phenomenon in optical systems. Magnetic lenses are an easy way to realize strong optical lenses and are typically used in many e-beam projection systems. Because they do not require any high voltage, but only a current generating the magnetic field, they are a simple and easy-to-control lens solution. However, due to the nature of how a magnetic field acts on a charged particle, magnetic lenses always act as a focusing lens, never as a divergent lens. In consequence, they alone cannot realize good aberration control with the method described in Fig. 4.

Fig. 4

Utilization of divergent electrostatic lenses allows us to realize a concept for aberration correction, commonly known in light optics, in the charged particle domain. By combining concave and convex lenses, this approach compensates for spherical aberrations and minimizes beam field distortion, two contributors for increasing blur and consequently limiting resolution.

Although the beam distortion in most electron optics applications only causes a displacement, in multi-beam applications, beam field distortion is a significant contribution to the total blur. Multi-beam technology makes use of massive averaging to limit imperfections of individual beamlets. This averaging overlaps different regions of the image field to write a single structure on the mask. By that, different regions of the image field are locally combined on the mask. With a larger residual distortion, the individual local displacements result in an additional blur component, as illustrated in Fig. 5.

Fig. 5

Effects of distortion of the beam array on the writing quality. On the top, representing concept A, a higher distortion is shown in red. When multiple regions of the beam array are overlapped to write a structure, as in multi-beam writing processes by, e.g., multi-passing or overlapping shots, this displacement between individual contributions results in a broader blur distribution, as illustrated with the envelope in green. Conversely, when the distortion is minimized, this effect is substantially reduced (concept B).

Upon inspecting overlapping passes closely, we find that registration remains surprisingly unaffected due to the center of the averaging. However, other parameters such as blur, dose slope, CDU, pattern-fidelity, resolution, and process window suffer from this higher distortion.

Controlling aberration and distortion blur together is the general challenge in multi-beam writing; together they define the electron optics baseline. On top, we need to add the Coulomb or space-charge related blur that is dependent on the current through the projection optics. It becomes evident that having a lower electron optics baseline allows a higher current in the projection system to be accepted without compromising the resolution, as illustrated in Fig. 6. The resolution, on the other hand, is the key to resolving small feature sizes and achieving small half-pitch dimensions as needed for the next generation EUV and high numerical aperture (High-NA) EUV masks.

Fig. 6

Comparison of blur performance with different concepts. The aberration correction and low distortion blur allow concept B to operate at a higher beam current to meet the same blur value.

In summary, concept B’s two-stage acceleration system ensures maximum APS health due to the absence of radiation and guarantees the lowest possible aberrations and distortion, thereby proving superior resolution and throughput.

4.1.

MBMW-100 Flex Enabling Multi-beam for Mature and Advanced Nodes Applications

There is a significant growth in chips in the mature nodes, mainly driven by the automotive industry, which requires highly efficient mask writers. Here, MBMW-100 Flex can deliver many advantages from multi-beam technology to applications with mature and intermediate nodes, allowing it to avoid running into the limitations of VSB technology. Based on the well-established MBMW-201 platform, MBMW-100 Flex was optimized for mature and intermediates node application requirements providing three different write modes to be applied for 32 nm node masks up to 10 nm node masks at write times of 3 to 5 h (mask writing field of $104\times 132\mathrm{m}{\mathrm{m}}^2$). The mask time varies with the chosen write mode, independent of pattern complexity.

This tool is designed for use with a wide range of different resists, from high sensitivity resists, starting at $10\mu \mathrm{C}/\mathrm{c}{\mathrm{m}}^2$, to medium and low sensitivity resists of up to around $50\mu \mathrm{C}/\mathrm{c}{\mathrm{m}}^2$. The newly developed electron beam source (eSource) system and write modes were specifically designed to meet these requirements. In addition to the new eSource, the Flex tool mostly builds on the hardware of the MBMW-201, with its demonstrated performance and durability over many years.

The MBMW-100 Flex was released in the second quarter of 2023, and the first unit was installed at the customer.

4.2.

MBMW-301 Redefining Leading Edge Mask Writer

The MBMW-301 is a revolutionary step in mask writer capabilities and comes with the next generation APS, a substantially smaller spot size, and a significantly faster data path. This technological advancement is required to meet the requirements for 2 nm node and below and High-NA EUV lithography,⁹ with regards to placement, CDU, and resolution for half pitch and assist features below 20 nm, while keeping the write time comparable to the predecessor.

With that, the MBMW-301 represents a revolutionary step for mask writing technology, embodying significant advancements over other mask writers. These enhancements promise to redefine industry benchmarks and ensure unmatched precision and efficiency in mask writing processes.

The MBMW-301 utilizes an evolved eSource, which increases the current density on the substrate by a factor of two compared with the MBMW-201. This higher current density in combination with the efficient writing strategy is essential to achieving the write time requirements on ultra-low sensitivity resists. The third generation APS controls close to 600,000 individual electron beamlets at a three times higher data rate compared with previous models. The tool’s air-bearing stage from JEOL allows for writing at double the stage speed at the required accuracy.

The MBMW-301 Alpha tool is currently in operation, and Beta tools are being shipped to clients. High volume manufacturing (HVM) units will be ready for deployment in the first quarter of 2024.