The semiconductor industry is the aggregate of companies engaged in the design and fabrication of semiconductors. A semiconductor is a material product usually comprised of silicon, which conducts electricity more than an insulator, such as glass, but less than a pure conductor, such as copper or aluminum. Semiconductors can be found in thousands of products such as computers, smartphones, appliances, gaming hardware, and medical equipment.
Type of Semi Conductors
- Memory: Memory chips serve as temporary storehouses of data and pass information to and from computer devices’ brains.
- Microprocessors: These are central processing units that contain the basic logic to perform tasks. Includes both central processing units (CPU) and graphics processing units (GPU).
- Commodity Integrated Circuit: Sometimes called “standard chips”, these are produced in huge batches for routine processing purposes. Simple chips used for performing repetitive processing routines. Produced in large batches, these chips are generally used in single-purpose appliances such as barcode scanners
- Complex SOC: “System on a Chip” is essentially all about the creation of an integrated circuit chip with an entire system’s capability on it. all of the electronic components needed for an entire system are built into a single chip. The capabilities of a SoC are more extensive than those of a microcontroller chip, which generally combines the CPU with RAM, ROM, and input/output (I/O). In a smartphone, the SoC might also integrate graphics, camera, and audio and video processing.
Automotive and Semiconductors
- Advances in technology such as AI, electric vehicles (EVs), autonomous driving, energy storage, and cyber security; social awareness of topics such as safety and ride-sharing; environmental concerns like pollution; and economic considerations including infrastructure spending and growth in Asian markets are all set to reshape the automotive industry
- Four mega trends will result in more semiconductor content being added to automotive electronics: Automation, electrification, connectivity and security
- Artificial intelligence framework can be broadly characterized into three layers
- Infrastructure layer includes the core AI chips and big data that support the sensing and cognitive computational capabilities of the technology layer
- The application level sits at the apex, providing services such as autonomous driving, smart robotics, smart security and virtual assistance
- AI chips form the heart of the AI technology chain and are central to the processing of AI algorithms, particularly for deep neural networks (DNN)
- CPU (Central Processing Unit): Traditional central processing units (CPUs) excel at general workloads, particularly if they are rules-based. However, CPUs can no longer keep up with the parallelism required of AI algorithms
- GPU (Graphic Processing Unit): used to process graphic intensive tasks such as games are built with parallelism in mind. GPUs have very high performance suitable
- for deep learning AI algorithms that require a lot of parallelism
- FPGA (Field Programmable Gate Arrays): programmable arrays suitable for clients that want to reprogram based on their own requirements. FPGAs are characterized by a faster development cycle (versus ASIC) and low power requirements (compared to GPUs).
- ASICs (Application Specific Integrated Circuits): dedicated architecture for AI applications. ASIC-based AI chips have many variations, including TPU, NPU, VPU and BPU, etc. These are all aimed at diverse, computer-intensive, rules based workloads with high efficiency and performance and the flexibility of a CPU. Typically, ASIC AI chips have higher efficiency, a smaller die size, as well as lower power consumption than GPUs and FPGAs. But, ASIC chips’ development cycle is longer and less flexible
- AI chip market is expected to account for over 12% of the total AI market by 2022, with a CAGR of 54%
- Network edge AI chips are emerging: AI chip deployment is not limited to the cloud, but can also be seen in a wide variety of network edge devices such as smartphones, autonomous vehicles and security cameras
- Inference AI chips in smartphones are now part of a three-way race between smartphone manufactures like Apple, Samsung and Huawei; independent chip providers such as Qualcomm and MediaTek; and IP license providers including ARM and Synopsys
- IP licensing vendors ARM and Cadence offer soft CPU and DSP IP cores, on the assumption that in the future, AI processing will be embedded in ASICs, rather than handled by stand-alone chips specialised for AI workloads. Their model allows silicon vendors to license the AI soft cores to develop their own chips targeting AI applications
Semiconductor Industry by Component Type
- Memory. A large portion of the growth in this segment will be driven by ongoing technological advancements such as cloud computing and virtual reality in end-devices such as smartphones. Sharply higher average selling prices (ASPs) for dynamic random access memory (DRAM) and NAND flash chips are also playing a significant role in generating revenues
- Logic. Demand from the communications, data processing and consumer electronics sectors will largely drive this market.
- Microcomponent. Automakers are incorporating them in large numbers into intelligent cars, for powertrains and next-generation chassis and safety systems, and to process sophisticated, real-time sensor functions in safety and crash avoidance systems.
- Analog. We expect strong growth fuelled primarily by demand from the communications industry and, especially, the automotive industry. Use cases generating growth in demand include power management (to increase cellphone battery life), signal conversion (for data converters, mixed-signal devices and others) and automotive-specific analog applications (autonomous and electric vehicles and electronic systems)
- OSD (optoelectronic, sensor and discrete components). solid-state lighting, machine vision, image recognition, smart-grid energy, IoT and multi-sensor ‘fusion’ in intelligent portable systems
Semiconductor Industry by Application Type
- Automotive. We expect the automotive market to grow the fastest of all the markets, with a CAGR of 11.9%. This is due largely to strong penetration rates of electric and hybrid cars, which require about twice the semiconductor content of conventional cars, and the strong market potential for autonomous driving
- Communications. Almost 80% of the demand for semiconductor from the communications market is driven by phones. Though the phone market is highly saturated, the introduction of 5G, the continuing high replacement rates of smartphones and the increasing demand for phones in emerging markets will maintain a CAGR of 2.2% for the market
- Consumer electronics. Semiconductor revenue from consumer electronics applications will be generated by TV devices, driven by the increasing popularity of smart TVs, 4K ultra-HD TVs, 3D programming, video-on-demand content, a preference for large displays, and curved OLEDs. Gaming technology and set-top boxes will also be strong revenue boosters. As a result, the market will grow at a CAGR of 2.2%. Although the wearables market is still relatively small, it will grow the fastest of all the consumer electronics applications, at a CAGR of 6.0%
- Data processing. Semiconductor sales in the data processing market, which includes devices such as PCs, ultra-mobiles, tablets, servers and storage devices, will grow at a moderate CAGR of 2.1% through 2022. A considerable portion of the market’s growth will come from storage devices, with a CAGR of 12.3%, as smart functions in end-devices require more semiconductor content
- Industrial. After automotive, the industrial market will grow the fastest among all application types; we expect a CAGR of 10.8% through 2022. The largest share of that growth will come from demand for security, automation, solid-state lighting and transportation. We expect demand for semiconductors for security applications to grow the fastest, at a CAGR of 17.8%
Semiconductor Industry by Application Type
- Automotive. These will include both inference-based systems, for self-driving and safety assistance in the car and at the edge, and training-based systems, for traffic avoidance mapping. The relative sizes of the two will determine the types of semiconductors that will witness the most growth in demand—GPUs and ASICs for edge computing and CPUs and FPGAs for cloud computing.
- Financial services. Use cases involving identity authentication for transactions and smart portfolio management. Authentication-based use cases will depend largely on inference-based AI on the edge, primarily for facial recognition on smartphones and fingerprint sensing through mobile CPU or dedicated AI semiconductors. Training-based AI will be used primarily to analyse massive data sets to recognize trends for smart investing and portfolio management; these activities will typically reside in the cloud, given their need for heavy computation based on CPU or GPU infrastructures.
- Industrials. Among all sectors, this one will likely present the smallest opportunity—between US$1.5bn to US$2bn—primarily from manufacturing optimisations and proactive fault detection. Realizing the benefits of AI will likely take this industry longer than other sectors, due to longer deployment and refresh cycles for industrial customers.
- Adoption of safety-related electronics systems has grown explosively. Semiconductor components that make up these electronic systems will cost USD600 per car by 2022. Advanced driver-assistance systems (ADAS) will have the largest increase
- The cloud is the biggest market for AI chips, as their adoption in data centers continues to increase as a means of enhancing efficiency and reducing operational cost
- Growth of the global semiconductor industry has been driven largely by demand from electronics such as smartphones and the proliferation of applications including the Internet of Things and cloud computing
- Head mounted displays will be the main driver of semiconductor growth in consumer electronics
- Wearables and smart watches will be new point of growth
- Data processing electronics include computing and storage. Storage, especially SSDs (solid-state drives), will account for the largest increase
- For accelerators, the training market is now evenly divided between CPUs and ASICs. In the future, however, we expect that ASICs built into systems on chips will account for 70 percent of demand. FPGAs will represent about 20 percent of demand and will be used for applications that require significant customizatiom
- When it comes to inference, most edge devices now rely on CPUs or ASICs, with a few applications—such as autonomous cars—requiring GPUs. By 2025, we expect that ASICs will account for about 70 percent of the edge inference market and GPUs 20 percent
- High-bandwidth memory (HBM). This technology allows AI applications to process large data sets at maximum speed while minimizing power requirements. It allows DL compute processors to access a threedimensional stack of memory through a fast connection called through-silicon via (TSV)
- On-chip memory. For a DL compute processor, storing and accessing data in DRAM or other outside memory sources can take 100 times more time than memory on the same chip
- One potential disruption in storage is new forms of nonvolatile memory (NVM). New forms of NVM have characteristics that fall between traditional memory, such as DRAM, and traditional storage, such as NAND flash. They can promise higher density than DRAM, better performance than NAND, and better power consumption than both
- Magnetoresistive random-access memory (MRAM) has the lowest latency for read and write, with greater than fiveyear data retention and excellent endurance. However, its capacity scaling is limited, making it a costly alternative that may be used for frequently accessed caches rather than a long-term dataretention solution
- Resistive random-access memory (ReRAM) could potentially scale vertically, giving it an advantage in scaling and cost, but it has slower latency and reduced endurance
- Phasechange memory (PCM) fits in between the two, with 3D XPoint being the most well-known example.