How Automation is Helping U.S. Manufacturers Scale Semiconductor Manufacturing

Semiconductors are core to all modern electronics, power distribution, and renewable power generation. Semiconductor products range from simple discrete components such as transistors and diodes to complex integrated circuits or ICs. Semiconductor devices are often at the core of logic gates that combine to make digital circuits. They’re also in oscillators, sensors, analog amplifiers, photovoltaic cells, LEDs, lasers, and power converters. Industry product categories include memory, logic, analog ICs, microprocessors, discrete power devices, and sensors.

Figure 1: The production of integrated circuits and other semiconductor products necessitate specialty equipment. (Image source: Getty Images)

Despite the critical nature of semiconductors, much of the world is dependent on undiversified and therefore vulnerable global supply chains. This is due to very significant economies of scale that make highly consolidated production more economically competitive. After all, semiconductor fabrication facilities cost billions to build and need very highly skilled staff.

Figure 2: Linear motors, belt drives, and miniature profile-rail linear guides are just some of the precision equipment in machinery to process semiconductors. (Image source: Getty Images)

Most fabs (foundries) are located in Taiwan, Japan, China, the U.S., and Germany and have been operating for decades. However, more than half of all semiconductors and more than 90% of all advanced semiconductors are made in Taiwan, with all major electronics manufacturers using a single Taiwanese semiconductor fabrication plant for at least some of their semiconductor fabrication. Recent geopolitical tensions have brought into sharp focus the dangers of such reliance. The 2022 Creating Helpful Incentives to Produce Semiconductors (CHIPS) and Science Act aims to address this issue by incentivizing operators and automation suppliers to establish and expand U.S. semiconductor production.

The state of semiconductor manufacturing

Most materials are either good conductors of electricity, such as metals, or insulators, such as glass. Semiconductors have electrical conductivity between that of conductors and insulators; that conductivity is adjusted by introducing impurities within the crystal structure via a process called doping. Doping with an electron-donor element gives a negative charge for an n-type semiconductor. Conversely, doping with an electron-acceptor element creates holes having positive charge for a p-type semiconductor. Two adjacent but differently doped regions within a single crystal form a semiconductor p-n junction. Transistors may be arranged with NPN or PNP junctions.

Silicon is by far the most common semiconductor material. Common n-type dopants are phosphorus and arsenic, while common p-type dopants are boron and gallium.

Figure 3: The six-axis robot in this Jabil Precision Automation Solutions machine executes tasks related to automated reticle sorting without compromising the contained cleanroom environment. (Image Source: Omron Automation Americas)

The most advanced semiconductor fabrication produces products having nanoscale features between 1 and 100 nm. As a nanometer is one billionth of a meter and the distance between individual atoms in a solid is between 0.1 and 0.4 nm, modern semiconductor nanostructures have approached the limit of how small material structures can be. The extreme precision involved in manufacturing such products demands processes executed in cleanroom environments as well as protected against vibration from seismic activity, local aircraft, trains, traffic, and nearby machinery.

The most significant processes in IC manufacturing are wafer production, lithography, and selective doping — most commonly by ion implantation. Many fabs specialize in either wafer manufacturing or the subsequent chip fabrication involving photolithography and doping. Taiwan Semiconductor (TSMC) produces both wafers and chips; it’s the only fab producing advanced 5-nm and 3-nm chips. Some semiconductor manufacturers such as Intel and Texas Instruments have their own fabs and only rely on TSMC to supply their most advanced chips. However, many fabless manufacturers (including Apple, ARM, and Nvidia) rely entirely on TSMC for their semiconductor fabrication.

Figure 4: GlobalFoundries recently began a $1B investment to allow its existing New York state facility to produce an additional 150,000 wafers per year. This new capacity aims to satisfy demand for feature-rich chips for automotive, 5G, and IoT applications. The facility will also support national security requirements for a secure supply chain. (Image source: GlobalFoundries)

While AMD is technically fabless, it isn’t reliant on TSMC and previously fabricated its own chips. AMD spun off its fabrication business and named it GlobalFoundries; the latter operates fabs in the U.S., Europe, and Singapore. Its New York fab historically produced chips down to 14 nm; on the horizon are 4-nm chips and then 3-nm chips.

Considering specific chipmaking processes

Much of semiconductor manufacturing employs scalable high-yield processes that allow the creation of millions of individual features (even nanoscale features) in a single step. Consider some of the specifics.

Silicon wafer manufacturing: Polycrystalline silicon nuggets are melted in a partially evacuated argon atmosphere and then pulled using a seed crystal to grow a single crystal silicon ingot — a cylinder having head and tail cones formed when the process is started and stopped. Some uniform doping may be added to the silicone at this stage.

Figure 5: Shown here several crystal silicon ingots and the disks that can be sliced from them. Cones are still present on the ingots after pulling and before grinding. (Image source: Getty Images)

Next, the ingot is ground into a block with a precise diameter and a notch is added to indicate the crystal orientation. The block is then sliced into wafers using a wire saw; wafers are beveled and lapped using diamond grinding tools; and then surface finishes are refined with chemical etching, heat treatment, polishing, and cleaning with ultrapure water and chemicals. Wafers are inspected for flatness and particle-free cleanliness before they are packaged.

Figure 6: Even seemingly familiar cleaning products take new forms when destined for use in cleanroom settings. (Image source: ACL Staticide Inc.)

Lithography: Electronic circuits are produced by first depositing a thin film of metallic conductor onto a semiconductor substrate and then using lithography to print a mask for the patterns of the circuit, before etching away the remaining conductive layer. These methods were originally developed for larger printed circuits but are now used for nanoscale fabrication of ICs. Metal fins are printed in a grid pattern, with 5-nm-process chips having fins spaced at a pitch of about 20 nm. Automated systems for this particular process often employ direct-drive technologies as well as stabilization bases and software and even air bearings.

Figure 7: Nanoscale structures can be investigated via electron microscopes as well as scanning tunnelling microscopes. Photomask repair equipment like that shown here automates defect detection and repair verification to accelerate throughput. Atomic force microscopy allows detection and repair of defects and foreign particles with nanometer accuracy and angstrom-level precision. (Image source: Park Systems)

Thin film material deposition: In this process, metallic material is deposited on the silicon wafer using vacuum evaporation, sputter deposition, or chemical vapor deposition.

Patterning: This is the actual lithography process during which the mask is applied to prevent the metal layer being removed from selected areas in the subsequent etching step. Common patterning processes include photolithography, electron-beam lithography, and nano-imprint lithography. Metal between the gaps in the mask is vaporized by a laser or electron beam.

Etching: The chemical removal of layers of material. Chemical wet etching uses reactive liquids such as acids, bases, and solvents, while dry etching uses reactive gases. Dry etching includes reactive ion etching and conductively coupled plasma etching. Here, automated equipment controls the process duration and rate — key to keeping chip features within tolerances.

Ion implantation: Once the grid of electrical connections has been created on a silicon wafer, individual transistors must be created at the junctions by doping the silicone to create NPN or PNP junctions. This is achieved by directing ion beams composed of the doping elements at the junctions. The very high velocity of accelerated ion beams causes them to penetrate the material and embed themselves in the crystal lattice of the silicon wafer. The patterns created during the lithography process are used to precisely guide the ion-implantation process.

Employing automation to deliver semiconductor quality

Much of the U.S. semiconductor industry currently produces fabrication equipment rather than actually fabricating semiconductors themselves. This equipment applies more conventional mechanical and electronics manufacturing automation technologies. For example:

• Lithography equipment is made by Applied Materials and ASML.
• Chemical-vapor deposition equipment is made by Lam Research and Applied Materials.
• Plasma etching equipment is made by Lam Research, Applied Materials, and Plasma-Therm.
• Ion-implantation equipment is made by Axcelis Technologies and Varian Semiconductor Equipment Associates.

Though currently the U.S. imports most of its semiconductor volumes, all stages of manufacturing are executed to some extent within the U.S. This includes both wafer and chip fabrication by Intel, GlobalFoundries, Texas Instruments, and others.

Processes for thin film material deposition, lithographic patterning, chemical etching, and ion implantation for chip fabrication are intrinsically scalable. They allow millions of individual junctions to be simultaneously created. Manufacturers are therefore increasing levels of automation in part to improve productivity — but more often these days to improve quality.

Automation is also associated with chemical, chip, and wafer-handling operations as well as the use of cleanroom robots produced by manufacturers such as KUKA Robotics. The latter play an important role in reducing losses caused by human errors.

Figure 8: Collaborative robots ride on seventh-axis systems to handle silicon wafers (40 µm thick and up to 300 mm in diameter) as they advance through up to 1,200 steps to be turned into chips. (Image source: KUKA Robotics)

But in semiconductor manufacturing, automation is often more about the processing of data and automation of resulting decisions. Fabs use automated algorithms for advanced process control or APC as well as statistical process control or SPC. These track process variations and resulting manufacturing defects to be reduced via real-time control over manufacturing processes. Such systems may employ artificial intelligence and machine learning to identify patterns within very large data sets tracking many process parameters and quality metrics.

Thought leadership at Siemens defines APC as encompassing various methods to reduce variation in control variables — including fuzzy control, model predicative control, model-based control, statistical model, and neural networks. Such Industry 4.0 technologies are often implemented via integrated ecosystems such as those offered by Siemens or Schneider Electric’s EcoStruxure (to give two examples) for the semiconductor industry. Process variables can be combined with machine condition monitoring for predictive maintenance that reduces routine production machine maintenance while avoiding downtime.

Conclusion

As the U.S. moves to ensure competitiveness of domestic production for strategically critical semiconductors, state-of-the-art automation will be essential. Cleanroom robots performing material handling are the most obvious and visible implementation of automation, but it’s the automated process control of the actual fabrication processes where real competitive advantages are gained. From controlling the environment for silicon crystal growth to ensuring precise doping at junctions during ion implantation, the efficient and defect free production of nanoscale ICs depends on the real-time control of thousands of process parameters.

Ultimately it will be advanced process control involving integration of IIoT sensors, AI algorithms, and other advanced model-based control methods that will ensure U.S. semiconductor-industry competitiveness.