MEMS, roadmap for the next ten years

The following is the text of the article:

As the 2D expansion of transistors slows down and 2.5D/3D packaging technology matures, the development of optical transceivers and interconnect devices that utilize integrated photonics technology, are manufactured in CMOS factories and adopt advanced integrated circuit packaging technology has begun to become a More important technological innovation. Optical transceivers co-located in the same package with data processing (computing) and memory chips are called co-packaged optics, or CPO. A chip containing active photonic devices and associated photonic circuits is called a PIC (Photonic Integrated Circuit). These PICs convert electrical signals into optical signals and transmit optical domain data between packages and/or between the computing core and memory within the package.

CPO's most important value propositions are improved bandwidth density and energy efficiency, two key metrics driven by growing demands for computing power and communications bandwidth. This need is particularly driven by the rapid growth and implementation of sophisticated AI and ML accelerators and computing clusters that are driving interconnect bandwidth, energy efficiency, and lower latency, from chip-scale systems to systems powered by thousands of Large-scale systems composed of GPU, CPU and memory ICs.

While co-packaging optical transceivers improves data transmission efficiency, optical sensors and actuators also play an increasingly important role in data collection and processing. These sensors and actuators are critical in products such as televisions, car headlights, projectors and DNA analysis chips, as well as in optical switches for data sensors and wearable heart rate and oxygen sensors. Many of these sensors are manufactured using microelectromechanical systems (MEMS) technology. MEMS devices and manufacturing technologies can also be used to adjust, modulate or modulate the arrangement of optical devices such as tunable filters, lasers and optical fibers, enabling new products such as near-infrared (NIR) materials analysis scanners.

In addition, the combination of MEMS and application-specific integrated circuits is also maturing at the same time, which will lead to unprecedented products.

The miniaturization of sensors has fueled the telemedicine revolution, allowing diagnostics to be performed in remote locations or at home, providing less invasive alternatives to surgery, and enabling implantable, ingestible or wearable sensors and neural probes. Tiny sensors also facilitate the use of personal digital twins so doctors can practice surgeries through simulations and provide more accurate visualization of the body, allowing for better diagnosis.

Sensors also enable innovation in smart homes, smart cities and advanced manufacturing facilities. New chemical sensors can detect gases, monitor pollution and air quality in buildings. The Internet of Things (IoT) is based on the availability of small-sized, low-power sensors. Sensors will also continue to find important applications in areas such as structural health monitoring and predictive maintenance, defense and aviation.

The market for handheld and wearable gadgets is expected to continue growing rapidly over the next decade. In addition, artificial intelligence-driven robots and self-driving cars will continue to be widely used. All of these applications require more sophisticated, reliable, lower-cost sensors with high-bandwidth interconnect capabilities.

Mobile phone economics are primarily driven by cost, size, performance and bandwidth. Robust sensor design is critical to the success of a phone's GPS, gyroscope, accelerometer, pressure sensor, magnetometer, optical image stabilizer, microphone and fingerprint sensing. Over the past five years, sensor sensitivity and accuracy have increased tenfold, while power consumption, cost and size have dropped by a fifth. These trends are expected to continue.

The fusion of physics and artificial intelligence in device computing enables better MEMS-based product design. The integration of these sensors enables activities such as navigation positioning, stability control, impact detection, adaptive lighting, image stabilization and traction control to be performed seamlessly. Better sensor performance means higher signal-to-noise ratio (SNR), higher dynamic range and sub-milliwatt power consumption.

Sensors with additional functionality are also needed. For example, silicon integrated components with a band gap smaller than silicon (Si) and capable of operating in short-wave infrared (SWIR) are needed to achieve high-resolution, eye-safe long-range lidar (LiDAR). It is also desirable to package these sensors in ultra-small packages. Flexible packaging is another emerging area that is important for wearable devices and medical applications.

Computing and intelligent processing close to sensors are critical for energy efficiency and latency, and co-optimization of hardware and software is an important vehicle for collaboration. TinyML (www.tinyml.org) is another rapidly growing area of ​​sensors and actuators. This area includes hardware, algorithms, and software near or on edge devices. For example, it might be on sensors at the edge of IoT data management and computing solutions. Typically, ML inference of sensor data is done locally and consumes about 1mW of power.

As TinyML proliferates and grows in the IoT over the next decade, key areas of focus will include low-power wake-up circuits, non-volatile memory integrated with silicon fabricated using advanced silicon nodes, and power generation at the milliwatt level. Power Budget Efficiently utilize limited memory/computing resources for ML algorithms.

Silicon photonics technology promises to expand frequency and bandwidth for a variety of applications including sensing, interconnection, communications and computing. Advanced optical sensors for health/medical sensing, including blood glucose detection, blood pressure, and heart disease markers, as well as sensors for automotive applications, such as lidar, present significant development opportunities. An important issue facing these sensors is how to achieve the accuracy and reliability of similar solutions.

To provide true solid-state beam steering for ADAS (Advanced Driver Assistance Systems) lidar systems, further development of optical phased arrays and gratings is required. Solid-state beam steering will reduce costs, increase reliability and shrink lidar scanning systems, all of which are necessary to enable mass production of SAE Level 3 and above autonomous vehicles.

ADAS lidar systems in mass-produced cars also require the integration of emitters with OPA (optical phased array)/grating elements and detectors with point cloud processing functions at the silicon level. Integration of these components is best achieved through code design of the components and wafer-level hybrid bonding or similar methods.

Systems that use sensor data often do not subject it to further security checks and trust the sensor data. Therefore, operating on the physical phenomena that the sensor is designed to account for may result in undesirable operating processes. For example, a MEMS accelerometer on a smartphone can measure steps by simply playing a YouTube video containing hard-to-hear sounds. You can even use this method to convey information. For example, by shining a laser on a window, criminals can inject commands into a voice-activated system.

Therefore, the security issue of integrated microelectromechanical systems (MEMS) devices is a challenge that needs to be solved to prevent malicious tampering of data. MEMS can also be part of a security solution, as MEMS devices can be used as part of the physical security mechanism that protects circuits from tampering.

Trends in sensor manufacturing and design include:

• Replace crystal oscillators with CMOS-compatible MEMS-based resonators to enable new architectures, higher performance, and elimination of off-chip passive components.

• Piezoelectric MEMS sensors and actuators are used in applications such as handheld ultrasound, as well as improved and miniaturized speakers and microphones.

• Employing a suite of technologies that leverage new low-cost materials and manufacturing techniques, lower-cost sensors can be produced in high volumes while producing high-precision sensors for mission-critical tasks such as GPS navigation.

• Installing sensors in clothing and fabrics creates new fashion categories that compete with sensors based on phones, rings, bodyguards and watches. Many innovative technologies also have military applications. Fabric sensors place new demands on connectivity, reliability and durability.

• Microelectromechanical systems drive advances in quantum computing because MEMS structures are used to enable qubits to communicate with the outside world.

Unlike many electronics, sensors are manufactured in a variety of processes, often targeting a specific application or sensor type. Depending on the sensor, it may need to be open to sense the surrounding environment while also preventing unwanted environmental effects. Some sensors are better suited for sealed packaging. Key factors in meeting the unique packaging needs of specific sensor structures include packaging of sensors fabricated using different manufacturing technologies and combining flexible and rigid sensors on flexible substrates. A competing trend is the standardization of sensor packaging solutions in certain applications, making the supply chain more efficient.

There are many ways to integrate sensors with associated electronics. Sensors can be built on the same chip as the electronics via special compatible processes, created on/under CMOS as a backend process, or combined as separate chips. Microelectromechanical systems (MEMS) can also be separated from the CMOS stack, and some researchers are even using fin structures as the basis for sensing elements. Separate chips or post-processing can use advanced CMOS nodes and optimized MEMS processes. These different integration strategies have implications for packaging, material selection, manufacturing, and assembly. Each of the above strategies will have its own advantages for the foreseeable future.

Challenges remain in integrating and power/area efficient optical-to-electrical-to-optical conversion and interfacing. To expand the application scope of this technology, these challenges must be addressed.

Integrated photonic technologies will greatly improve communications infrastructure. It is expected that within the next five years, channel symbol rates will easily exceed 100 Gigabit baud, and the aggregate bandwidth of integrated photonics transceiver modules for fiber optic communications will reach 3.2 to 6.4 Tbits/s. The urgent need for increased bandwidth and energy efficiency will drive the transition of data center architecture from pluggable optics to co-packaged optics (CPO), thereby driving the transition from copper to shorter distances (sub-1 meter) fiber optics.

It must be emphasized that the enormous bandwidth of optical fiber cannot be fully exploited without using dense wavelength division multiplexing (DWDM, or similar multiplexing methods) to place multiple optical channels on a single fiber (bandwidth density achieved through photonic technology). order of magnitude increase). Successfully implementing DWDM or coherent technology on an application-specific integrated circuit also allows for a trade-off between channel data rate and number of channels for a given fiber bandwidth. This trade-off can improve energy efficiency and reduce system cost. In the coming years, as the bandwidth density of photonic interconnects increases and their applications become more widespread, reducing the energy consumption, cost per bit and latency of the entire optical system will become even more important, especially for artificial intelligence/mobile computing applications. Packaging optics close to ASICs and other computing integrated circuits using short, low-loss channels will help improve some of the energy efficiency and bandwidth, enabling low-power electro-optical conversion and high-bandwidth data transmission in large-scale systems.

Currently, analog photonic links are used to simplify millimeter-wave node integrated circuit architecture, allowing the number of antennas per chip to exceed 1,000. Analog photonic links also greatly improve energy savings in the process and are far more energy efficient than digital links in the signal chain.

When it comes to communications applications, there is still a large amount of the electromagnetic spectrum that remains untapped. Taking full advantage of this vast untapped spectrum requires technological innovation. Innovative semiconductor technologies, such as SOI/SiGe-based photonics, VCSELs, micro-LEDs, avalanche photodiodes, and InP-based PICs, will provide advanced process platforms that will enable many future improvements in bandwidth and power.

In the field of MEMS, optical switches are replacing OEO switches. MEMS-based tunable filters and capacitors, RF switches, and the aforementioned MEMS resonators are enabling new architectures and higher levels of integration, resulting in reduced footprint and simplified packaging. These new components usher in a new era in communications circuit and system design. Leveraging MEMS enables tunability, alignment, and calibration of communication components and improves specifications.

Photonics offers significant opportunities to address power and bandwidth bottlenecks in high-performance and data center computing systems as data movement continues to expand. New high-speed, low-power transceivers, light sources, waveguides/modulators and photodetectors are some of the fundamental building blocks of integrated photonics - an area where sustainable dimensional scalability remains a key challenge. Photonic technologies offer opportunities for low-power, high-speed I/O and photonic interconnect structures. Additionally, photonic devices have been shown to perform certain mathematical operations, such as matrix-vector multiplication.

By enabling photonic integration near the processor (CPU/GPU/FPGA/ASIC) and in the link from the processor to memory, the bandwidth density can be reduced by taking advantage of the huge bandwidth and low loss advantages of optical transmission compared to copper transmission. >100x improvement, >10x improvement in energy efficiency, and extended package-level performance system-wide.

Optical links provide energy-efficient, low-latency interconnects that enable the disaggregation of networks, compute and memory. This will leverage optics co-packaged with ASICs/CPUs/GPUs and use optical communications standards to connect compute and memory across racks in hyperscale data centers.

Artificial intelligence accelerators and neuromorphic computing are other exciting areas driving higher compression ratios, which, combined with advanced 2.5D/3D package integration, have huge potential for improved energy efficiency. Light-based AI accelerators are currently being explored, and increased investment in this area may place additional demands on the PIC roadmap. Investment in research and development of silicon photonics technologies for quantum computing is also increasing, which requires new materials and processes.

MEMS-based products require more on-board computing to produce smarter sensors. MEMS manufacturers have "moved up the food chain" and are no longer limited to providing raw sensor output. Today, typical MEMS sensors provide intelligent motion or other processed data. This is part of the smart edge processing trend. The architecture that combines computing and sensing is changing rapidly, impacting CMOS-MEMS integration and advanced packaging.

To achieve higher performance and higher integration density, innovative semiconductor process platforms include SOI/Ge-based photonics; integrating III-V materials through epitaxial growth, wafer/die bonding, or laser intracavity connections; On silicon wafers for light sources, modulators and detectors; and active devices based on physical effects other than plasmon dispersion effects (plasmonics, graphene).

Lithium niobate and barium titanate films for hybrid integration are enablers for high-frequency modulation, while photonic wirebonding for laser-encapsulated interconnects using 3D printing technology are other areas requiring further research. High bandwidth, low transmission loss and low power chip-to-chip optical interconnects are also expected to require advances in embedded waveguides in substrates/PCBs. It is important to ensure low-cost, high-reliability photonic materials that maintain stable performance over long periods of time and over a wide temperature range, while also maintaining low thermal hysteresis and low loss characteristics, thereby requiring low overall energy consumption.

For some sensors and actuators, CMOS compatible scaling is required while increasing inertial sensor performance and RF filter power handling capabilities, so new materials such as tungsten or other high atomic mass metals need to be explored to reduce the overall footprint . In addition, aluminum nitride-based CMOS-compatible platforms are also being explored. In addition, new piezoelectric and phase change memory sensors with higher force density and linearity are also being investigated. These new materials for sensors and actuators will in turn lead to innovations in packaging materials and technologies that require temperature and humidity control.

Two important material drivers for inertial sensors are the ability to prevent the mechanical device from sticking to the substrate, and a material set with a good CTE match to prevent unnecessary bending and stress, especially those generated by packaging. These sensors also require improved low-cost acquisition materials and coatings that resist wear, adhesion, corrosion, and charge elimination. For chemical sensors, the material requirement is to facilitate chemical reactions in a repeatable and stable manner. Some requirements for acoustic sensor/actuator materials such as microphones/speakers are to facilitate controlled motion to generate, move and detect sound.

Because these devices must withstand the large number of cycles, a key feature of MEMS materials is that they deform predictably and do not fatigue. Optical sensor/actuator materials include those that produce flat, reflective surfaces that are resistant to deformation and have appropriate optical properties. In RF MEMS switches, the development of reliable contact materials remains very important.

In photonics and microelectromechanical systems applications, there are several emerging next-generation materials that deserve attention and should be considered for mass production. Precursor characterization efforts are needed to integrate these materials into semiconductor manufacturing and to develop automated front-end manufacturing equipment to apply or remove these materials. Table 6.1 below lists some examples.

Efforts to introduce photonics technology into integrated circuit packaging have already begun. In recent years, some initial products showing the transition from stand-alone transceivers to CPOs or optical engines designed for CPOs have been demonstrated.

Figures 6.1 and 6.2 illustrate these early CPO products. Figure 6.1 shows the Ranovus Odin optical transceiver, operating at 896Gbps, integrated on the same package substrate as AMD/Xilinx's Versal FPGA. It was demonstrated at the 2022 Optical Fiber Communications (OFC) conference. Figure 6.2 shows the integration of Ayar Labs' TeraPHY optical transceiver with a data processing integrated circuit. These TeraPHYs use 8λ wavelength division multiplexing technology (WDM) and are supported by the 8λ light source - SuperNova.

The most advanced MEMS products combine multiple sensors with electronics to provide advanced outputs processed by onboard low-power electronics that often integrate artificial intelligence and advanced calibration capabilities. Products from Bosch, STMicroelectronics, Invensense, Analog Devices, Texas Instruments and other companies all come with APIs and can be plugged into the system at any time. Emerging MEMS products incorporate advances in sensor/actuator manufacturing technology, materials and design. Here are some examples of emerging products:

• Microelectromechanical system-based speakers

• Chemical sensors that can sense a variety of substances

• Ultrasound array for handheld imaging instruments

• BAW devices that can be integrated with CMOS

Figure 6.3 illustrates state-of-the-art MEMS products. Shown in the figure is eXo Imaging's low-voltage pMUT array chip: Exo Silicon. Exo Silicon combines the proven imaging performance of piezoelectric crystals with the economics of silicon. Each chip contains 4096 independently controlled pMUTs with large bandwidth, unparalleled sensitivity and an ultra-wide field of view of up to 150 degrees. Exo's silicon architecture enables rapid improvements in imaging quality and provides real-time artificial intelligence capabilities that analyze every frame of image to guide users to immediate answers. Going forward, pMUT technology will enable powerful 3D imaging and potentially 4D imaging. Nursing staff will be able to observe patients better and make diagnoses faster.

Optical communications are nothing new. More than half a century ago, with the maturation of optical fiber manufacturing technology and the emergence of optical transmitters and detectors based on III-V technology, the role of optical transceivers in long-distance data transmission rapidly expanded - thanks to glass fiber Low loss, low dispersion and wide bandwidth characteristics. Advances and diversification of technology have led to significant cost reductions, resulting in the widespread use of optical transceivers in access and client applications.

At the same time, the emergence of DWDM technology and erbium-doped fiber amplifiers (EDFA) in the 1990s revolutionized long-distance networks and laid the foundation for the information superhighway. Over the years, optical transceivers have evolved from custom-designed devices, to on-board modules, to a host of multi-source protocol specified small form factor (SFF) pluggable optical transceivers. These independent pluggable transceivers play an integral role in today's data centers.

However, as data volumes and the computing power of data processing integrated circuits continue to grow, stand-alone transceivers will soon become the bottleneck for high-bandwidth data transmission. The total number of transceivers installed on the panel may not provide sufficient bandwidth for the data processing units or data switches on the line cards. As data transfer rates increase to 100 Gigabaud or higher, copper wire losses between the integrated circuit's serial data switch and the optical transceiver data input pins will become a challenge from a signal integrity perspective. Therefore, CPO provides a good alternative in this scenario and a compelling option to break down this physical barrier while continuing to reduce overall system power consumption and cost.

Placing optical transceivers in the same package solves signal integrity issues by eliminating long copper traces on line cards with shorter, lower-loss interconnects and potentially direct-drive optics. However, to truly overcome the bandwidth density bottleneck, we may need to introduce DWDM technology for long-haul/long-haul transceivers while maintaining the short-haul same-package optical Ioss mechanism.

Because MEMS devices often require custom processes, it is difficult to create a highly standardized platform like CMOS. For example, magnetic sensors may require materials that inertial sensors do not. Some manufacturers attempt to offer manufacturing processes that make multiple types of sensors on the same chip, but cost and performance drivers often dictate custom processes. Therefore, the MEMS device manufacturing process and its associated packaging must be coded and designed.

Another limiting factor is that there are no primitive components like CMOS, namely transistors. Test methods are often unique to the sensor’s operating principle or application, so the test and assembly infrastructure, ecosystem, and supply chain are more complex than for electronics. Although great progress has been made in these fields, these problems still limit the widespread application of MEMS technology and make the time to market of MEMS-based products an issue. As the MEMS market continues to grow rapidly, more suppliers will launch MEMS products and the situation will improve.

Communications, computing and memory applications need to address technical and supply chain challenges in multiple areas:

• Photonic IC (PIC) Performance (200G/Lambda+), Yield, Manufacturability and Cost

• Mass production of low-cost DWDM III-V laser sources for silicon photonics (especially O and C bands)

• DFB (Distributed FeedBack) QDOT and other laser purchasing options, performance, plug efficiency and cost

• Unsealed lasers offer high performance at elevated ambient temperatures and are suitable for packaging within ICs

• Integrated high-power laser for high output and reliability

• Hybrid integration of laser materials

• Edge-Coupled and Vertical-Coupled Fiber Connectivity Passive/Active Solutions, Fiber Pitch Scaling and Cost

• Optical fiber assembly process - low coupling loss, high throughput, high yield

• Fiber ribbonization of fiber arrays with high fiber count/density and reduced cladding diameter

• Fiber array interconnect/termination hardware development and standardization

• Fiber Management for Co-Packaged Optics (CPO) – High Density Fiber Per Package (High Fiber Count) and Ribbon Fiber

• Advanced heterogeneous packaging, including 3D TSVs enabling high bandwidth density, high SI (signal integrity) and low interconnect power consumption

• Fiber optic bus architecture for processor-processor and processor-memory access

• Low-cost light sources for optical interconnects • Thermal tunability and junction temperature management • Design for test (DFT) and design for manufacturability (DFM)

• Stable and viable PIC, laser and optical fiber integrated supply chain and ecosystem support

• Photonic circuit modeling standard for ecosystem support

Sensor/actuator applications require solving technical and supply chain challenges in multiple areas:

1. CAD

• Nonlinear reduced-order modeling of sensors/actuators

• MEMS co-design (sensors/electronics and packaging)

• PDK (Process Design Kit) with material properties in all relevant physical areas

2. Materials

• Characterization of new materials in all relevant fields of physics

• Material synthesis tools to discover and optimize materials with desired properties

• Characterization of materials under bending and tension, especially for wearable devices

3. Standards

• Material bending and tensile properties standards

• Sensor PerformanceFOM Standard

• Reliability and testing standards for emerging technologies

4. Workforce Development

• Training of students in multiple physical fields required for MEMS (e.g. mechanics and electronics)

• Develop training to more fully engage bachelor's and master's students in MEMS and photonics design, as has been done for VLSI

5. CMOS and multi-sensor integration

• The transition from stacked wire-bonded sensors must continue using new packaging methods to enable greater heterogeneous integration

6. Sensor design and manufacturing improvements

• Improve inertial sensors to navigation grade by using field calibration, multiple sensors, and/or combination with other non-MEMS sensors

• Improved design and manufacturing methods, and process window enhancements to compensate for manufacturing non-idealities

• MEMS-based energy harvesters must increase sensor conversion power output percentage to compete with solar and thermoelectric devices

• Continued development of low- and near-zero power sensors to meet energy needs

• Optical glucose sensors must become more accurate to compete with needle-based electrochemical sensors

• Accuracy of paper and plastic sensors must improve to compete with silicon-based sensors

• Continued research into atomic clock technology to replace large components

*Disclaimer: This article is original by the author. The content of the article is the personal opinion of the author. The reprinting by Semiconductor Industry Watch is only to convey a different point of view. It does not mean that Semiconductor Industry Watch agrees or supports the view. If you have any objections, please contact Semiconductor Industry Watch.

Today is the 3601st issue shared by "Semiconductor Industry Observation" with you. Welcome to pay attention.