Chip Industry Week In Review

By Susan Rambo, Gregory Haley, Jesse Allen, and Liz Allan

President Biden issued an executive order on the “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.” It says entities need to report large-scale computing clusters and the total computing power available, including “any model that was trained using a quantity of computing power greater than 1,026 integer or floating-point operations, or using primarily biological sequence data and using a quantity of computing power greater than 1,023 integer or floating-point operations.” Also on the must-report list is “any computing cluster that has a set of machines physically co-located in a single datacenter, transitively connected by data center networking of over 100 Gbit/s, and having a theoretical maximum computing capacity of 1,020 integer or floating-point operations per second for training AI.”

In response to the executive order, the Department of Commerce (DOC) said experts at the National Institute of Standards and Technology (NIST) will lead technical work on AI safety. NIST, meanwhile, called for participants in a new consortium. In addition, the Department of Energy (DOE) detailed initiatives it will lead, and the Department of Homeland Security (DHS) explained its critical role.

The UK hosted a global AI Safety Summit, bringing together governments, companies, civil society groups, and experts to consider AI risks and how they can be mitigated through coordinated action. Speaking to the BBC at the event, Elon Musk said the technology poses a pressing existential risk.

The regulations are still being set, but the chip industry is already in motion. Arm is working with AMD, Intel, Meta, Microsoft, NVIDIA, and Qualcomm on initiatives focused on enabling advanced AI capabilities for more responsive and more secure user experiences. Arm and NVIDIA are adapting NVIDIA TAO, a low-code open-source AI toolkit for Ethos-U NPUs. Arm and Meta are bringing PyTorch to Arm-based mobile and embedded platforms at the edge with ExecuTorch.

In addition, Cadence announced Voltus InsightAI, a generative AI tool that automatically identifies the root cause of EM-IR drop violations early in the design process and selects and implements fixes to improve power, performance, and area (PPA).

And Tignis outlined ways that AI and machine Learning (ML) can help solve the chip industry worker shortage – not by replacing people, but by helping teams become more efficient.

Quick links to more news:

Design and Power
Manufacturing and Test
Automotive
Security
Pervasive Computing, AI, and Connectivity
Top In-Depth Reports
Events

Design and Power

Ansys announced a new voltage drop prevention, analysis, and sign-off technology called Sigma DVD that it says answers the longstanding challenge of predicting dynamic voltage drop (DVD). This is particularly acute at the most advanced nodes, where DVD can account for a 10% drop in achievable F_max, or the inability to meet V_min. The problem is that it’s nearly impossible to capture billions of possible switching combinations that can lead to voltage drop using traditional vector approaches.

Accellera made its Clock Domain Crossing (CDC) Draft Standard 0.1 available for public review. The effort aims to ease SoC integration by defining a standard CDC collateral specification that every tool can translate their native format to and from.

NVIDIA created a domain-adapted large language model (LLM) for chip design. Researchers from the company explored using custom tokenizers, domain-adaptive continued pre-training, supervised fine-tuning with domain-specific instructions, and domain-adapted retrieval models to customize a foundation model. That model then was used for an engineering assistant chatbot, EDA script generation, and bug summarization and analysis.

Renesas uncorked a new series of microcontrollers that include Arm Cortex-M85 processors. Renesas said the RA8 devices deliver 6.39 CoreMark/MHz performance and use Arm’s M-Profile Vector Extension to boost performance for DSP and ML. The family targets edge and endpoint devices that implement natural language processing in voice AI and predictive maintenance applications.

Synopsys debuted the Synopsys Cloud OpenLink program to enable chip designers to access third-party EDA tools and IP in its cloud environment. Synopsys is releasing an API program that members can use to transfer entitlements to Synopsys Cloud.

Imperas Software announced several product updates, including the latest models of RISC-V processors, ImperasDV processor verification solutions, virtual platform based tools for software development and architecture exploration, and the free RISC-V instruction set simulator. Imperas also teamed up with Tenstorrent to provide a model of the Tenstorrent Ascalon RISC-V core.

Arm acquired a minority stake in Raspberry Pi, a key platform for low-cost IoT development. Arm said the goal is to lower the barriers to innovation.

Realtek used the Cadence Tempus Timing Solution to sign off an N12 high-performance CPU core, citing improved PPA.

Spirent Communications and Cadence are bringing virtual Ethernet traffic emulation and testing capabilities to pre-silicon verification in the Cadence Palladium Z2 Enterprise Emulation and Protium X2 Enterprise Prototyping systems. The solution has the capacity to emulate any port speed from 1G to 800G at the application level.

Arteris and Semidynamics are partnering to support interoperability between Semidynamics’ Atrevido and Avispado 64-bit RISC-V processor IP cores and Arteris’ Ncore cache coherent network-on-chip (NoC) system IP for AI/ML and HPC designs.

Gowin Semiconductor expanded its Arora V high-performance FPGA family.

IAR added support for Renesas RA8 MCUs in the latest release of IAR Embedded Workbench for Arm.

Manufacturing and Test

UMC initiated the wafer-to-wafer (W2W) 3D-IC project, in collaboration with Winbond, Faraday, ASE, and Cadence, to help customers accelerate production of 3D products. The project offers an end-to-end solution for integrating memory and processor with silicon stacking to target edge AI applications. The platform is expected to be ready in 2024.

U.S. wafer start capacity may reach 46 million in 2027, according to TECHCET, a 35% jump from the 34 million starts expected for 2023, due to fab expansions coming online over the next four years. The domestic demand for bulk chemicals is expected to increase by a combined 75% over that time, creating a gap in the supply needed to support the industry.

Jabil will take over the manufacture and sale of Intel’s current silicon photonics-based pluggable optical transceiver product lines, and the development of future generations of such modules.

SEMI announced the availability of the Industry 4.0 Readiness Assessment Model (IRAM) to help organizations across the semiconductor supply chain assess and track smart manufacturing technology deployment progress, and to develop a roadmap for their digital transformations.

Powerchip plans to bring a new fab online in 2026 for the 28, 40, and 55nm process nodes. The fab will be located in northern Japan’s Miyagi prefecture. The new fab is being built in cooperation with SBI Holdings (acting jointly under the entity name JSMC Co., Ltd), with an initial investment of about 400 billion yen (~US$2.67 billion).

JCET will build an advanced packaging facility for automotive chip products in the Lingang Special Area of Shanghai. The company is set to receive a capital increase of RMB 4.8 billion (~US$656 million) from multiple industry funds for construction.

Global semiconductor sales rose 1.9% sequentially in September compared with August 2023, but dropped 4.5% compared with sales in September 2022, according to the Semiconductor Industry Association (SIA). The good news is sales are 6.3% higher in Q3 2023 than Q2 2023, but they are still 4.5% lower than Q3 2022.

Meanwhile, worldwide silicon wafer shipments declined 9.6% in Q3, a 19.5% drop from the same period in 2022, according to the SEMI Silicon Manufacturers Group (SMG).

Digital transformation (DX) — the effort to enhance or convert manual processes into digital — is picking up steam, says IDC. The firm forecasts DX spending will reach nearly $3.9 trillion in 2027, with a five-year CAGR of 16.1%.

Malaysia is looking for foreign investors for its domestic EV and chip industries, according to Reuters.

Dutch chip companies may be investing in Vietnam in the near future, according to Reuters. The BE Semiconductor Industries (Besi) will invest $5 million to rent a factory in the country.

A China-backed chip investment fund invested 14.56 billion yuan ($1.99 billion) in a memory chip company called Changxin Xinqiao, according to Reuters.

Astera Labs selected proteanTecs’ health and performance on-chip monitoring solutions to enable ML-driven diagnostics and fleet management capabilities for multiple data center applications.

Automotive

Tesla won its jury trial in Riverside County Superior Court, California, over whether its autopilot technology was responsible for a crash that killed a driver in 2019. The company faces ongoing probes, investigations, and lawsuits concerning other autopilot incidents. 

Toyota will invest nearly $8 billion more into its North Carolina battery plant, bringing the total to about $13.9 billion, to add capacity for battery electric vehicles (BEVs), plug-in hybrid EVs (PHEVs), and hybrid EVs (HEVs). Line launches are planned through 2030 to reach a total production of over 30 GWh annually.

Volvo launched an energy solutions business unit, offering energy storage and charging-related technologies and services such as bi-directional charging to allow an EV to give back extra battery power to a compatible grid. The fully electric EX90 SUV will be the first Volvo model equipped with hardware and software to enable bi-directional charging and direct energy storage from solar. Meanwhile, Volvo’s CEO said the company will delay looking for a fourth EV battery factory site amid slow demand.

Subaru will adopt Tesla’s North American Charging Standard (NACS) for the charging ports of its BEVs to be launched in North America beginning in 2025, and will continue to advance NACS adoption for BEVs launched after that.

Rivian announced its “customers will soon begin receiving vehicles that have had their first charge entirely from wind and solar energy,” in a post on X (formerly Twitter).

In Australia, Freight and rail company Wabtec and iron mining company Roy Hill debuted a 100% battery-powered, heavy-haul locomotive for mainline service, featuring an energy capacity of 7 megawatt hours (MWh). “By using regenerative braking, it will charge its battery on the 344 kilometer (214 mile) downhill run from our mine to port facility and use that stored energy to return to the mine, starting the cycle all over again.”

Infineon and Eatron partnered to bring advanced machine learning solutions and algorithms to the AURIX TC4x microcontroller (MCU), aiming to advance automotive battery management systems (BMS).

A connected and automated vehicle (CAV) traveling on a multi-lane highway with integrated traffic light timing control can maximize energy efficiency and achieve up to 27% savings, according to a demonstration by Oak Ridge National Laboratory (ORNL) researchers.

Security

The Securities and Exchange Commission (SEC) charged SolarWinds and its CISO with defrauding investors and customers by concealing its poor cybersecurity practices. “The true state of SolarWinds’ cybersecurity practices, controls, and risks ultimately came to light only following a massive cyberattack.” In 2020, the Sunburst incident impacted SolarWinds’ Orion software platform, which accounted for 45% of its revenue that year.

The Common Weakness Enumeration (CWE) posted Version 4.13 with one new weakness entry and changes to 148 CWEs. Demonstrative examples were added to 56 CWEs, including nine that provide valuable illustrations of specific hardware weaknesses including Verilog hardware description language for fictitious system on chip (SoC) devices.

A new architecture and verification approach for hardware security modules was introduced by researchers at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and New York University.

Codasip announced the first commercial implementation of an advanced security mechanism called Capability Hardware Enhanced RISC Instructions (CHERI), enabling secure-by-design products so companies can take preventive security measures without waiting for vendors’ delivered patches. CHERI was developed at the University of Cambridge and proven in experimental processors.

Synopsys published its annual “Open Source Security and Risk Analysis” (OSSRA) report, examining vulnerabilities and license conflicts found in about 1,700 codebases across 17 industries, and offering recommendations for security, legal, risk, and development teams to better understand the security and risk landscape of open source development and use.

The Cybersecurity and Infrastructure Security Agency (CISA) issued numerous alerts including an advisory for Mitsubishi Electric’s CNC Series devices, which “could allow a malicious remote attacker to cause a denial-of-service condition and execute malicious code on the product by sending specially crafted packets.” CISA also launched its Critical Infrastructure Security and Resilience Month.

Pervasive Computing, AI, and Connectivity

SiTime plans to acquire the clock products of Aura Semiconductor. The all-cash transaction includes fixed payments totaling $148 million, with earn-outs capped at $120 million. The deal is expected to close by the end of 2023.

Western Digital plans to separate its HDD and flash memory business, creating two independent public companies. The move comes after merger talks between Western Digital and Kioxia were suspended.

Voltage Park launched its AI cloud service to make machine learning (ML) infrastructure more accessible. With about 24,000 NVIDIA H100 GPUs, its clusters consist of 80GB H100 SXM5 GPUs fully interconnected with 3.2T InfiniBand. Bare-metal access is currently available to be followed by short-term leases and hourly billing.

The UK’s fastest AI supercomputer, Dawn, was deployed and is now ready to use. Dawn was built by the University of Cambridge, UK Research & Innovation, the UK Atomic Energy Authority, Intel, and Dell, bringing the UK closer to reaching the compute threshold of a quintillion floating point operations per second.

The University of Bristol was awarded £225 million (~$273 million) by the UK government to build an AI supercomputer. Isambard-AI will be built by Hewlett Packard Enterprise and contain 5,448 NVIDIA GH200 chips.

MIT and NVIDIA researchers developed techniques that accelerate the processing of sparse tensors, which could result in improved performance and energy efficiency of systems like the massive ML models that drive generative AI.

The Federal Communications Commission (FCC) authorized over $18 billion to expand rural broadband.

U.S. smartphone shipments declined 19% in Q3 2023, the fourth straight quarter they have dropped, according to Counterpoint. Samsung, Google, and TCL saw the steepest declines (26%, 37%, and 51% respectively), while Apple’s shipments were down just 11%, partly due to the later launch date of the iPhone 15. Motorola and Nokia increased their shipments by 31% and 17%, respectively.

The Open RAN market will stagnate in 2023 and 2024, according to Counterpoint. Investments will start to increase YoY after 2025 with network operators investing a cumulative total of more than $30 billion between 2022 and 2030, representing 24% CAGR from 2023 to 2030.

Flex Logix’s EFLX eFPGA is being adopted for a high-volume data center application.

Infineon and DH-Robotics introduced a series of next-generation electric grippers, ideal for industries that require handling delicate components with accuracy. The grippers bring together MOSFETs, integrated industrial microcontrollers, high-precision Hall effect sensors, gate driver ICs, and PSoC 6 microcontroller units (MCU).

Siemens Digital Industries Software and Ricoh are collaborating on an industrial aluminum Binder Jetting (BJT) solution for mass production, with Ricoh leveraging Siemens’ Additive Manufacturing Network capabilities to maximize efficiency and achieve scale.

Winbond held its 2023 Carbon Neutral Family Day centered on the theme “Embracing AI for a Sustainable Green Future,” and unveiled its corporate song.

Top In-Depth Reports

Advanced node chips were expected to be prohibitively expensive to develop. That hasn’t turned out to be the case. So what happened?

The lines between counterfeiting, security, and design flaws are becoming increasingly difficult to determine in advanced packages and process nodes.

Evolving from hardware-based automotive design to a software-based, software-defined vehicle architecture is a complicated process.

What’s needed to secure data across multiple chiplets and heterogeneous systems.

Why a shortage of data often impedes root-cause failure analysis.

Events

Find upcoming chip industry events here, including:

Upcoming webinars are here.

Further Reading and Newsletters

Read the latest special reports and top stories, or check out the latest newsletters:

Automotive, Security and Pervasive Computing
Systems and Design
Low Power-High Performance
Test, Measurement and Analytics
Manufacturing, Packaging and Materials