XPENG completes tape-out of self-developed TURING chip

Shanghai (Gasgoo)- On August 27, XPENG celebrated its 10th anniversary in Beijing where it announced its strategic vision for the next decade, aiming to become a global AI automotive company. The event also marked the official market launch of XPENG's intelligent, all-electric hatchback, the MONA M03, with deliveries set to begin in September.

During the event, XPENG provided a first look at its AI ecosystem, which is built on a foundation of high-power computing chips and advanced AI large models. The XPENG AI automotive framework serves as the core, with AI-powered robots and flying cars as key branches. Moving forward, XPENG plans to fully integrate AI into its product offerings, establishing itself as a company that develops and implements both software and hardware in-house.

A major highlight of the event was the unveiling of "XPENG TURING," the world's first chip designed for simultaneous application in robots, AI cars, and flying vehicles. This groundbreaking chip successfully completed its tape-out process on August 23.

Named in tribute to Alan Mathison Turing, the "father of computer science" and "father of artificial intelligence," the XPENG TURING chip is tailored for end-to-end AI models. It features two proprietary neural network processing units (NPUs), two independent image signal processors (ISPs), and employs a domain-specific architecture (DSA) optimized for neural networks. With 40 cores, the XPENG TURING chip delivers triple the computational power of existing chips and can locally run AI models with up to 30 billion parameters. To meet the safety requirements of Level 4 autonomous driving, the chip also includes an independent safety island capable of real-time, blind-spot-free safety monitoring.

In XPENG's AI strategy, chips function as the "physical brain," while end-to-end models represent the "thought process." As of May this year, XPENG became one of only two automakers globally to mass-produce vehicles equipped with end-to-end AI models. The company introduced a new "End-to-End Four-Step" plan, aiming to deliver a Level 3+ user experience using Level 2 hardware and costs.

XPENG also revealed its next-generation "AI Eagle Eye Vision Solution," specifically designed for end-to-end AI. According to the company, this advanced system, powered by mass-produced end-to-end AI models, enhances visual information processing efficiency by eight times over its predecessor, improves overall computational efficiency by 20%, and reduces response time by 100 milliseconds, offering a smart driving experience comparable to that of LiDAR systems.

On the hardware front, the "AI Eagle Eye Vision Solution" integrates two 8-megapixel front and rear cameras, significantly upgrading precision, distance, and color resolution. The Lofic architecture significantly improves backlight detection capabilities, ensuring clearer vision in low light, backlight, and high-contrast environments. On the software side, the end-to-end AI model eliminates the need for data conversion, allowing visual information to be directly input into the neural network, thereby enhancing sensitivity and response time in perception and control.

XPENG has also developed its intelligent driving software for compatibility, ensuring that either vehicles equipped with the AI Eagle Eye Vision Solution or the LiDAR-based Max models will receive synchronized upgrades to maintain a consistent high-level smart driving experience.

The XPENG P7+ will be the first model equipped with the "AI Eagle Eye Vision Solution" in the fourth quarter of this year, with additional models, including those in the MONA series, to follow.

In addition, XPENG's flying car, dubbed the "Land Aircraft Carrier," will enter pre-sales in the fourth quarter of this year, and XPENG’s second-generation humanoid robot will debut at the "1024 Tech Day" event.