AI has undeniably become the leading force in today’s market. With the rapid rise of AI technologies, companies focused on AI chips, algorithms, and applications have emerged as key players. Many of them have already gained recognition from both customers and investors due to their unique strengths. Kneron is one such company that stands out.
After securing over $10 million in Series A funding from notable investors including Alibaba, Qijing Optoelectronics, China Development Capital, Qualcomm, Zhongke Chuangda, Sequoia Capital, and Entrepreneurship in November last year, Kneron recently launched new AI IP products. In a competitive landscape filled with both domestic and international rivals—such as ARM and Cambrian—each has its own advantages. ARM benefits from a vast ecosystem encompassing CPUs, GPUs, tools, and fabrication support, while Cambrian found its foothold through the adoption of Kirin 970 in smartphones.
In a market where AI giants already dominate, how can a startup like Kneron break through and survive? To understand this, semiconductor industry observers conducted an exclusive interview with Dr. Liu Juncheng, the founder of Kneron, to explore his strategy for "breaking the siege."
Starting with customization, focusing on NPU
Before founding Kneron, Dr. Liu Juncheng worked at companies like Samsung and Qualcomm. In 2014, driven by the potential of AI NPU (Neural Processing Unit), he left Qualcomm after working on neural network terminal applications for more than three years and founded Kneron.
An NPU is a specialized chip designed for deep learning, operating independently from the traditional von Neumann architecture. It is specifically tailored for AI tasks, making it one of the most effective ways to implement artificial intelligence today. Huawei’s Kirin 970 uses an NPU, but other options like CPU, GPU, FPGA, and DSP also exist. However, Kneron has focused on developing IP products for AI-specific chips, operating under ARM's IP licensing model.
When asked why he chose this path, Dr. Liu explained: “Choosing IP reduces cost and risk, and offers flexibility. Developing chips directly requires significant capital, which is a heavy burden for startups. The IP licensing model allows us to collaborate with partners, helping promote AI more widely.â€
With the team’s efforts, Kneron launched its first IP in 2016, primarily customized for specific customer needs. For example, they partnered with Tencent to develop intelligent navigation, face recognition, and AI algorithms, even contributing to Tencent’s “God Eyes†smart car project.
After gaining extensive experience and understanding customer needs, Kneron launched three major product lines.
Pushing Three Major Product Lines, Targeting the Lowest Power Consumption of 5mW
This time, Kneron introduced the Kneron NPU IP series.
According to Kneron, the Kneron NPU IP is a dedicated AI processor for edge devices, enabling offline execution of deep learning networks like ResNet and YOLO. This complete edge AI hardware solution includes hardware IP, compiler, and model compression, supporting various mainstream CNN models such as ResNet-18, ResNet-34, VGG16, GoogleNet, and LeNet, as well as popular deep learning frameworks like Caffe, Keras, and TensorFlow.
Combined with Kneron’s image recognition software, the Kneron NPU provides real-time analysis and fast response, ensuring stability and meeting security and privacy requirements. Due to tight hardware-software integration, the overall solution is compact and low-power, aiding rapid product development.
Kneron now offers three versions: ultra-low power KDP300, standard KDP500, and high-performance KDP700.
The KDP300 targets 3D face recognition in smartphones, using dual-lens structured light and stereo imaging for faster and more accurate identification. It also supports ultra-low power consumption, with less than 5mW for computation and SRAM.
The KDP500 enables fast real-time facial, gesture, and body recognition, suitable for smart homes and security systems. It delivers up to 152 GOPS at 500MHz with only 100mW power consumption.
The KDP700 handles advanced AI operations and deep learning inference, applicable to high-end smartphones, robots, drones, and surveillance devices. It offers up to 4.4 TOPS at 1GHz with 300–500mW power consumption.
Dr. Liu noted that these products are the first to achieve sub-5mW power consumption and millimeter-scale dimensions, allowing for compact device designs.
“Thanks to technical improvements, our performance per watt has significantly increased. Previously, it was around 1–1.5 TOPS/W, but the KDP700 reaches up to 11.6 TOPS/W using a 16nm process,†he emphasized.
The KDP500 is expected to enter mass production in Q2, with KDP300 and KDP700 also in discussions with customers.
Where Are the Masters?
Kneron is not the only player in the NPU space. Cambrian, another major player, has already been validated through the Kirin 970. As a relatively new entrant, how does Kneron compete? Dr. Liu didn’t answer directly but stressed that AI solutions—whether software or hardware IP—are application-dependent. There are clear differences between solutions for smartphones and those for vehicles.
He highlighted several advantages that position Kneron for future success:
First, Kneron offers a comprehensive product line with excellent specifications. The KDP300, for instance, can perform 3D image analysis and recognition in smartphones with ultra-low power, something no other manufacturer currently achieves.
Second, compared to vendors that offer only hardware architecture IP, Kneron’s NPU IP includes model compression technology, which can reduce model size by dozens of times. It also includes pre-trained data, making it a full software-hardware solution.
Third, Kneron has developed its own image recognition software, which integrates tightly with the NPU. This results in a smaller, lower-power solution that helps customers develop products quickly.
Additionally, Kneron’s wide range of products gives it a strong advantage over many competitors.
Market competition is fierce, but niche markets are the key to success.
No matter how lucrative a market may be, it always comes with high risks, and the AI chip market is no exception. From the beginning, Kneron has faced intense pressure and challenges. However, according to Dr. Liu, the real challenge isn't from specific competitors but from the ever-changing AI market itself, which is still very new and rapidly evolving.
“In order to find a place in this market, finding a niche is crucial,†Dr. Liu told reporters. Kneron has been proactive in this regard, having started early in the AI field and continuously accelerating its pace. This allowed them to develop products and gain customers before other companies even began discussing their AI strategies.
Taking the smartphone market as an example, Kneron has long recognized it as a key battlefield for AI applications. Knowing there are many competitors, including ARM, they planned ahead, offering a complete solution across high, medium, and low-end product lines.
According to Gartner, although AI on smartphones is still in its early stages, by 2022, 80% of smartphones will have built-in AI features. This aligns with Kneron’s “niche†strategy.
Dr. Liu emphasized that the market is full of possibilities and uncertainties. Those who can lead and respond quickly will gain an advantage.
“Our goal is to make AI accessible to everyone, not just high-end products,†he said. He added that in the AI market, no single player can succeed in all areas, but a dominant player will emerge in a specific segment.
When asked if Kneron has a chance to become the ARM of AI IP, Dr. Liu acknowledged that achieving such influence would require a broad industry presence, such as in cars and smartphones. While Kneron’s automotive layout is not yet deep, it has a strong position in smartphones, especially this year, which could bring great opportunities.
Beyond smartphones, security and IoT are also key areas, particularly IoT, where Kneron sees few competitors.
Dr. Liu stated that if the company can achieve tens of millions or even hundreds of millions of shipments in these three areas, its influence will be undeniable. He believes that one or two of these areas have a good chance of reaching such numbers.
In this way, becoming the ARM of the AI industry is not just a dream—it’s a realistic goal.
High Voltage Resistant Wire,High Voltage Wire,High Temperature Resistant Wire,Silicone Rubber Cable
JIANGSU PENGSHEN HIGH TEMPERATURE WIRE CABLE CO., LTD. , https://www.pengshencable.com