Gasgoo Munich-"On paper, today’s chips look impressive. CPU and NPU specs keep climbing, yet finding the right fit for actual needs is tough." At The 6th Automotive Semiconductor Ecosystem Conference 2026 hosted by Gasgoo, Ding Ke, senior chief engineer of chip development at Changan Automobile, captured the dilemma automakers face when selecting components. Specs don’t automatically translate to experience. Delivering on that promise requires both vehicle-level balance and a smoothly connected supply chain.
Bridging the gap between a spec sheet and a mass-produced car involves multiple hurdles: translating requirements, fusing architectures, and passing certification. The main battleground for automotive chips is shifting from a contest of device-level parameters to a race for system-level ecosystem synergy.
Who Should Define the Automotive Chip?
The difficulty Ding Ke describes highlights a disconnect in the definition phase: the peak performance listed on a spec sheet doesn’t always match a vehicle’s actual requirements.
Bridging this gap requires "translation": converting an automaker’s demands for safety and experience into precise chip-level specifications.
"You must truly understand the needs and failure mechanisms behind the specifications," said Xu Yudong, vice president of Chipsea Technology and president of its automotive electronics unit. Defining automotive chip specs—whether for information security or reliability—cannot stop at migrating industry standards. Chipmakers need to proactively step forward and co-develop with automakers and Tier 1 suppliers. "Even if it’s just a single IP, it needs to be jointly defined."
Once specs are defined, another translation is needed: conveying scenario requirements to the foundry and converting them into process parameters. Vehicle experience, product specifications, and process parameters—these three languages must connect seamlessly. Any break in the chain diminishes the final result.
Chipmakers cannot pull this off alone. Automakers and Tier 1 suppliers hold the first-hand data on vehicle scenarios and failure boundaries, while chipmakers must translate those demands into achievable specs and feed constraints back. Requirements are rarely nailed down in one go; they take multiple rounds of joint definition to converge.
Xu also cautioned that small chips outside the spotlight are just as critical. "Big chips set the ceiling for the vehicle experience, but small chips are the floor." Large chips determine the upper limit of intelligent experience, while MCUs and analog signal chain chips scattered across powertrains, chassis, and engines determine whether the car can run safely and reliably at all.
Foundries at the top of the chain are redefining their roles too. Guo Ru, head of GlobalFoundries' automotive business, admitted that the chip shortage five years ago exacted a heavy price: when capacity tilted toward industries with larger volumes and higher value, automotive clients were devastated, and some emerging chip companies didn't survive.
Realizing they couldn’t just stick to contract manufacturing, GlobalFoundries launched a division dedicated to the automotive market: one end secures capacity and supply safety, the other pushes technical alignment far upstream. Wafer fabrication is a long-cycle industry where process iterations take years. Roadmaps for automakers and chipmakers must be aligned early.
Image source: GlobalFoundries website
"Since the chip crunch, automakers, chipmakers, and foundries are tighter than ever," Ding Ke observed. For a chip to supply well, it must be good from start to finish: well-defined, well-designed, and built on solid foundry devices. Only then can you achieve overall optimization.
Yet the cost for automakers to switch is exorbitant. Whether swapping a chip or a model version, collaboration is required right down to the compiler level. Specs may shine individually, but the experience must be a shared responsibility across the chain.
The goal of collaboration is to minimize late-stage rework. Ding Ke put it bluntly: when automakers, chipmakers, and foundries connect the entire chain, end users are willing to pay, and every player in the chain earns a reasonable return.
As for how to speed up that "translation," Xu pointed to AI. In the era of physical AI, model-based systems engineering promises to make converting requirements more efficient. China’s automotive chip industry now has an opportunity to accelerate this translation process, ensuring defined products truly hit customer pain points.
Cabin-Driving Fusion: How to Clear This Hurdle?
The quality of this collaborative definition is being tested by cabin-driving fusion. Single-chip cross-domain solutions have already entered mass production, but Black Sesame Intelligent’s assessment is relatively sober: while industry leaders are blazing the trail, the massive base is following slowly. "I wouldn't say it's mainstream yet, but it certainly fits the market trend."
This trend already has production samples. Black Sesame’s Wudang C1296, the first locally produced cabin-driving integrated chip, is installed in Dongfeng Motor’s Tianyuan Smart Cockpit Plus platform. It is moving from small-batch designation to mass installation in the second half of the year.
Behind this trend lies practical calculation: when different schemes and configurations require different chips, supply chain management and development coordination become incredibly complex. If one chip can balance cockpit and intelligent driving needs, "everyone saves trouble, the supply chain is easier to manage, and all parties can make money."
However, high-end models will firmly move toward centralization, though the final form is far from locked in. Black Sesame product expert Erite pointed to the mature mobile phone industry as a reference: as phone makers enter the fray, they influence the existing market格局. The future may not be dominated by a single architecture, but rather by multiple parallel technological paths.
For chipmakers, the task is to support the complex, diverse configurations demanded by automakers at the lowest possible cost. That is why Black Sesame is currently focusing on volume models, leaving higher-end cross-domain and central computing platforms for the right moment.
The trend is clear, but implementation is not easy. He recalled that it took over three years from the company’s cross-domain chip landing to actually being installed in a car. The delay wasn’t in functional development itself, but in structural adjustment: cockpit and intelligent driving are separate departments within automakers. When two functions are squeezed into one chip, the question of "who takes the lead on development?" becomes a major resistance.
In his view, chipmakers better understand what algorithms the market needs, while automakers better understand what functions are required. Early communication isn’t hard; the difficulty lies in the vast amount of unmastered know-how between communication and implementation. Moreover, the automotive chain is long, and a single product iteration often takes years—the pace simply can’t be fast.
There is still no standard answer for the boundaries of cabin-driving fusion. In terms of standard-setting, the definition of the "cabin-driving integrated" controller is blurring: what level of intelligent driving counts as cabin-driving? How much cockpit integration is the optimal solution?
The real challenge isn’t stacking computing power, but grasping the precision of overall product definition. How far to integrate and how to achieve the optimal solution—automakers and chip companies are still weighing these trade-offs repeatedly.
As architectures centralize, the first thing to change is computing power distribution. Xu Yudong observed that as computing power moves upward, product forms at the edge are also shifting. Some scenarios no longer require traditional mid-to-low-end MCUs, but high-performance, high-functional-safety MCUs are still essential for critical roles in safety control and execution.
Chipsea’s positioning is no longer limited to categories like MCU, ASIC, or analog devices. Instead, it returns to a fundamental question: what is the need behind the need? How can subsystem functions be optimally combined, and how should edge execution and edge safety be understood?
"Have we really understood the safety requirements?" In his view, software doesn’t necessarily have to run on an MCU; it can move up the stack. In some scenarios, ASICs can replace MCU functions. But in others—especially those tied to safety and reliability—ASIL-D MCUs, or even just a small core, must be retained for sensing and safety execution.
Image source: Chipsea Technology website
Forms can change in a thousand ways, but the prerequisite remains thoroughly understanding the safety solutions that have persisted in engines and chassis for years—and figuring out which functions move up and which move down.
Advanced-node big SoCs attract the most attention, but beyond them lies a vast array of peripheral devices.
Guo Ru introduced that GlobalFoundries currently mass-produces down to 12 nanometers. While it can’t make cabin-driving SoCs, it is deeply involved in supporting chips for radar, image sensors, high-speed interfaces, and gateways—mostly handled by mature and specialty processes. Applications like driver monitoring and using large models to predict battery life are driving the integration of analog and digital processes on a single chip, merging signal sampling with computation. Technologies and customers from the robotics and server sectors are also intersecting more with automotive.
Viewed differently, AI is not the patent of the most advanced processes. Even non-leading processes like 7-nanometer or 10-nanometer have their own AI missions.
Among all players, automakers are the most eager for integration, yet Changan remains cautious about implementing cabin-driving fusion. The contradiction lies here: intelligent driving demands stability, while cockpits need to "show off." The two temperaments naturally clash.
Single-chip solutions hold advantages in cost and supply chain management, but they impose higher requirements for functional isolation, fail-safe degradation, and software complexity. This is precisely why automakers remain restrained in the face of integration.
The automaker’s ultimate pursuit is to pull LiDAR point cloud processing, ISP, and everything else into the central brain, achieving full-chain integration from light detection to brake output. This remains difficult to achieve. The rumored "three days per minor version" update cycle is just as hard to realize at the scale of mass integration.
For a time, cabin-driving fusion products may resemble "semi-finished" goods: the broader the integration, the higher the cost-performance ratio, but development itself is endless. The nuance required is far beyond what a single chip can achieve overnight.
This also means that centralization and decentralization are not mutually exclusive alternatives, but a re-division of labor based on functional safety levels. Computing power suitable for moving up centralizes, while links involving safety and real-time response firmly remain at the edge.
Cars Are Speeding Up—How Can Chips Keep Pace?
The final architecture is unclear, and all definitions and integration must ultimately withstand the long-term test of mass production. The sharpest contradiction is rhythm.
Vehicle model iterations can no longer return to a five-year cycle. A two-year cycle already stresses many automakers, and functional updates every six months are the norm. Yet a chipmaker’s product validation cycle is far longer than a vehicle iteration, and foundry process evolution is measured in years. This misalignment between fast and slow constitutes the third challenge of system-level competition.
"Every user likes fast iterations and hopes the things they buy will have extended lives—or even enjoy a second spring," Ding Ke said. Continuous evolution is a legitimate user expectation and a consumer mindset cultivated by the mobile phone and internet industries. Automakers must respond to it.
But the particularity of automobiles lies in life safety. Any part related to safety must complete all testing and certification—no shortcuts, no matter how long the cycle. This is the bottom line. Functions unrelated to safety, such as cloud features, can be updated rapidly at internet speed, giving users continuous new experiences.
The prerequisite for running these two rhythms in parallel is an excellent software architecture. Only by clearly delineating boundaries can automakers truly be responsible to users.
Chip companies are exploring methodologies for "front-loading" and "reusing" verification. Xu Yudong admitted that while iteration is a necessity, the rhythm is not uncontrollable. He offered two key approaches.
First is front-loaded verification. Clients often say they have no needs or resources to invest, but once needs arrive, they want a year’s worth of work delivered in six months. POC (Proof of Concept) verification makes things happen early, trading upfront investment for less trouble later on.
He also called for a breakthrough in mechanisms. He hopes that key specifications and safety attributes verified in early test chips can be partially accepted in subsequent ABCD sample verification and even chip replacement, turning upfront investment into an industry asset.
Second is platformization. Based on AEC-Q100 certification, first push the platform’s "ceiling" to the limit, use certification results as a general data reference, and then derive standardized configurations. "80% platform, 20% customization"—this is his formula for MCU products, aimed at significantly accelerating subsequent iterations.
Black Sesame’s strategy was summarized by Erite in four words: small steps, fast walking. Each step isn’t large, but the frequency is high enough to keep up with the market.
Its cross-domain chips share an almost identical product network with the previous generation of intelligent driving chips. Verified functions on driving chips can be quickly ported to cross-domain chips. Their new CPU and GPU architectures extend to the next generation of driving products, while the cross-domain chip itself validates multi-IP integration.
The two product sequences connect and extend each other, saving development costs for Tier 1 suppliers and automakers, while allowing products to move naturally from cross-domain to central computing, winning a longer life cycle.
The slowest and most costly link remains wafer fabrication. Guo Ru offered a longer timeline: moving from 40 nanometers to 28/22 nanometers can take three to five years of R&D. Adding automotive certification adds another one or two years. This means foundries must plan platforms for 2029 or 2030 this year. A misjudgment means massive investment goes down the drain.
Facing the supernormal pace of the Chinese market, the industry has also found a compromise path: first launch with preliminary automotive-grade processes—not fully automotive-grade, but with automotive screening added in packaging and testing—then gradually reinforce automotive safety certification on the back end, running multiple lines in parallel.
It must be clarified that this compromise is mostly used for components with lower safety levels. For safety-critical links like powertrains, chassis, and intelligent driving, complete automotive-grade processes and certification remain a non-negotiable prerequisite.
Meanwhile, custom chips are becoming a new variable. International leaders like Tesla and domestic head automakers are all stepping in to define proprietary chips, while Tier 1 suppliers like Bosch and Continental are also joining the customization trend. What foundries can do is keep an eye on all links of the chain and allocate resources based on return on investment.
For independent chip companies, this customization trend is both a new opportunity to win orders and a potential long-term threat to the market space of standard products.
Fast and slow are not opposites. Safety-related links cannot be compressed, but experience-level functions can iterate rapidly. Front-loaded verification, platform reuse, sequence continuity, and process pre-research—all are ways to use controllable preparation today to handle future demand uncertainty.
As automotive chip competition extends from spec sheets to overall vehicle definition, whether the numbers on a single chip look pretty is no longer the sole factor for success. How early automakers, chip companies, and foundries align their needs and match their rhythms is becoming increasingly critical. There are no lone winners in system-level competition; the depth of collaboration across the upstream and downstream supply chain will directly influence the speed of China’s automotive chip iteration.
