Panther Lake, Intel’s Core Ultra Series 3 processor family, is more than just a new generation of laptop processors. It is also the company’s first product to be manufactured at high volume using the Intel 18A process. As such, it represents a major test of Intel’s ability to return its manufacturing technology to the forefront of the semiconductor industry—and eventually offer it to external foundry customers as well.

In a new 90-minute documentary released by Intel, the program’s managers and engineers describe a development process that began in 2021 and involved thousands of employees around the world. While the film was produced by Intel and frames the project as a success story, it also provides a revealing look at the challenges of developing a processor and the manufacturing process on which it depends at essentially the same time.

Intel’s Israeli development center played a central role in the project. According to the documentary, the product definition process began in a room at the company’s Israeli R&D center, where teams tried to anticipate what the PC market would look like two or three years into the future. The compute tile and CPU cores were developed in Israel, while the graphics technology was developed by teams in Folsom and India. Other components came from additional Intel sites worldwide.

The Israeli executives and engineers featured in the documentary include Zohar Tzaba, vice president and Panther Lake program manager; Yaron Ananki, vice president and head of the Client Development Group; Erik Gihon, senior principal engineer for SoC architecture; Hagai Burstein, product development engineering manager; and Liel Elgarably, client platform integration engineer.

Lunar Lake Efficiency, Arrow Lake Performance

The architectural starting point for Panther Lake was an effort to combine the strengths of Intel’s two preceding processor families.

“Panther Lake is a combination of Lunar Lake and Arrow Lake. We took the best of both worlds and brought them together,” Erik Gihon explains in the film.

From Lunar Lake, Intel carried over power efficiency, battery life and the NPU. From Arrow Lake, it drew performance, scalability and a higher core count. In its largest configuration, Panther Lake features up to 16 CPU cores, compared with eight in Lunar Lake, along with integrated graphics containing as many as 12 Xe cores. Its new performance cores are called Cougar Cove, while the efficiency cores are based on the Darkmont architecture. Both were adapted for the Intel 18A process.

To address a broad range of PC designs, Intel moved from a monolithic architecture to a modular one. The family includes CPU, graphics and I/O tiles of varying sizes that can be combined in three main configurations. An eight-core version is aimed primarily at business laptops. A 16-core configuration with a larger I/O tile adds eight PCIe 5.0 lanes and is designed for PCs equipped with discrete graphics. The flagship version combines the larger processor with integrated graphics featuring 12 Xe cores.

A Manufacturing Process Still in Motion

The project’s unusual challenge stemmed from Panther Lake’s role as the “lead customer” for Intel Foundry’s 18A process. Unlike a product developed on a mature manufacturing node, Panther Lake was designed using evolving versions of the process design kit and frequently updated transistor-performance models.

The product teams had to anticipate how the manufacturing process would behave once it reached maturity and continually adjust the design accordingly.

Intel 18A introduces two major technological changes: RibbonFET, Intel’s gate-all-around transistor architecture, and PowerVia, a backside power-delivery system. In the documentary, Bill Grimm, product engineering manager in Intel Foundry’s Technology Development organization, describes 18A as “a revolutionary process, rather than simply an incremental improvement over the previous generation.”

PowerVia brings power connections closer to the transistors instead of routing them through 14 or 15 metal layers on the front side of the chip. The result is lower resistance, reduced area requirements and potential improvements in both performance and energy efficiency.

The change, however, also forced Intel’s engineers to redesign some of the signal-routing infrastructure. The front-side power-delivery network had previously provided shielding between sensitive signals. Once it was moved to the back of the chip, some metal lines had to be widened and spaced farther apart to prevent interference and preserve clock-signal integrity.

Making Progress With Imperfect Chips

Before the first silicon arrived, Intel tested the design, firmware and software using simulations, emulators and FPGA-based systems. It then began a gradual engineering-sample process. The first ES0 units initially operated at low frequencies. Performance improved through the ES1 and ES2 generations, eventually leading to the B0 revision, which was selected as the production candidate.

Because yields on the new manufacturing process were still limited, Intel developed a method for making use of units that were not fully functional. If only one section of a chip was defective, the unit could be assigned to a validation team that did not require that particular component. This allowed Intel to extract value from nearly every unit coming out of the fab and continue testing despite the limited supply of fully functional samples.

The modular architecture added another layer of complexity. Each processor combines CPU, graphics and I/O tiles produced on different wafers. It was therefore not enough to sort and test each tile separately; Intel also had to match the tiles in combinations that would produce a functioning processor at the required performance tier.

The Bet on B0

One of the project’s most consequential decisions came roughly six months before product qualification. Intel needed to begin building commercial inventory so that hundreds of thousands—and eventually millions—of units would be available at launch. At that point, however, the company still did not know with certainty whether the B0 revision was free of defects that might require another silicon stepping.

The decision was based on the progress of validation, the number of units tested, the status of known defects, yield forecasts and feedback from PC manufacturers. Program executives say it also involved an element of engineering judgment, drawing on the experience of the architects, developers and validation teams.

The gamble paid off: B0 became the production revision of Panther Lake.

At the same time, Intel was working with PC manufacturers on approximately 200 different system designs.

“If all 200 systems had been powered on for the first time at once, our validation and customer engineering teams would have been completely overwhelmed,” says Liel Elgarably.

Intel initially selected a small group of systems and brought them up and validated them together with the manufacturers. Only after resolving the first round of issues did it expand the process to the remaining designs.

The company also dramatically shortened the product-qualification stage. While qualifying the various Alder Lake packages took more than a quarter, Intel had to qualify the entire Panther Lake family within a single month. Findings from the first product were shared daily with the teams responsible for the other configurations, helping them avoid repeating the same tests and encountering the same problems.

For Intel, Panther Lake is therefore more than a new processor. It is the company’s first attempt to turn RibbonFET and PowerVia from laboratory-developed technologies into the foundation of a product family manufactured in tens of millions of units.

The documentary, however, does not disclose yield figures, rejection rates or manufacturing costs. The project’s success will ultimately be measured not only by the fact that Panther Lake reached the market, but also by its real-world performance, the availability of PCs based on the processor, and Intel’s ability to attract external foundry customers to its 18A process.

Watch the full documentary:

How do you defend an organization in the AI era? In a special TechTime series, senior leaders from the cyber industry set out the new threats for CEOs and CISOs – and the defensive approach now required.

Guy Tytunovich, founder and CEO of CHEQ

The entire world of identity was built on a single foundational assumption: that on the other side of the interaction is a human being. Usernames, passwords, and two-factor authentication were all designed to verify that a person is who they claim to be. That assumption is collapsing before our eyes.

We are entering an era in which a vast share of digital interactions will be carried out by AI agents acting on behalf of humans – searching, comparing prices, filling in forms, and even making purchases. The implication is that organizations need to learn not just who is in front of them, but what is in front of them.

“Who are you?” is no longer enough

Identity management has to change at its root. Instead of asking only “who are you?”, it needs to answer three questions in real time: is the entity in front of me a human, a bot, or an AI agent; on whose behalf is it acting; and what is its intent?

Agents themselves also need an identity and permissions model, with a clear chain of delegation: who dispatched them, what they are allowed to do, and what reputation they have built up over time.

An organization that can’t tell the difference between a legitimate agent buying on behalf of a real customer and a hostile bot impersonating one will lose twice: the customer, and the security battle.

The threat is at the front door

Security teams today know very well what is happening inside the corporate network, but they are almost blind to what is happening on the digital front line – on websites, in apps, in forms and in APIs.

Think about a bank robbery: the robber comes in through the branch, the place where the money meets the general public. He doesn’t head for the corporate offices of the company that owns the bank. An organization’s digital branch is its website and its app — and that is where many of the new threats operate.

The method of detection has to change too. An attack generated using AI can be polymorphic — every instance of it looks different. So we need to move from detection based only on signatures and rules to detection of behavior and intent.

And when the attacker is autonomous, the response also has to arrive at machine speed. A model in which a human analyst approves every action simply doesn’t survive the pace of events. The SOC of the coming years will run more autonomous defensive agents, and the analyst will shift gradually from being the person putting out fires to the person managing the system.

The danger: a reality that doesn’t exist

The threat that, in my view, gets talked about far too little is actually data contamination. Not one big attack, but a quiet flooding of the internet with synthetic entities.

Armies of bots and agents can open accounts, fill in forms and generate leads, clicks and engagement that look entirely human. The result is that the CRM fills up with people who don’t exist, the analytics describe an audience that doesn’t exist, and business decisions get made on the basis of a fabricated reality. Worse still, the organization’s own AI may end up training on data that was itself created by another AI.

The more organizations open the door to legitimate agents acting on behalf of customers, the greater the incentive for attackers to impersonate precisely those agents.

What should you do tomorrow morning?

If I were stepping into the CISO role tomorrow at a company with 5,000 employees, then before budgets and tools I would map the organization’s digital front door: websites, apps, forms and APIs.

I would ask a simple question: how much of the traffic there is human, how much is automated, and how much of it is already coming from AI agents.

You can’t protect what you can’t see. In the new world, organizations will have to treat non-human entities as full citizens of the identity system too — with permissions, accountability, and the ability to know who sent them and what they are allowed to do.

The next stage is already visible on the horizon: agent authentication. Just as SSL became a standard without which a browser won’t trust a website, we will reach a world in which a business can’t trust an agent without knowing who it is and on whose behalf it is acting.

 

image above: An illustration of a digital model of a wave energy generation system. Source: EWP.

Eco Wave Power Global (EWP) announced that its wholly owned U.S. subsidiary, Eco Wave Power U.S, has entered into an agreement with Germany-based AI engineering GmbH to develop a physics- and data-driven Digital Twin platform for Eco Wave Power’s proprietary wave energy technology. Both Eco Wave Power U.S. and AI engineering are members of the NVIDIA Inception program.

Under the agreement, the companies have commenced Phase 1 of the project, combining advanced physics-based simulation with EWP’s real-world engineering, wave and operational data. The initial phase will focus on digitally modeling how ocean waves interact with EWP’s proprietary floaters. Using AI engineering’s PAMICS simulation technology, the companies will evaluate floater behavior, structural loads and theoretical energy input under different sea conditions.

“For us, AI is an opportunity to fundamentally improve how we understand, design, predict and ultimately operate wave energy systems,” said Inna Braverman, Founder and CEO of EWP. “Phase 1 of this collaboration is therefore very practical. We will model wave loading, floater behavior, structural forces and theoretical energy input, while beginning to develop machine-learning capabilities for forecasting loads and energy yield. Importantly, we will also examine how those models can be transferred to new sites with different wave conditions.”

Digital representation of EWP system

AI engineering is specializes in applied fluid mechanics, numerical flow simulation and high-performance computing. Its PAMICS technology is a particle-based multiphysics simulation framework designed for complex fluid-dynamics and fluid-structure interaction applications. “Wave energy is a particularly challenging engineering problem because it combines complex free-surface fluid dynamics, structural motion, and highly variable environmental conditions,” said Dr. Stefan Adami, Co-Founder and General Manager of AI engineering GmbH.

“Our goal is to create a physics-based digital representation of Eco Wave Power’s system that links high-fidelity simulation with operational data. This approach allows us to better understand energy capture, structural loading, and system behavior across a wide range of sea states. Combining these physics models with ML can support more accurate forecasting, design optimization, and scalable deployment of wave energy technology.”