At the recently held ICDIA, Deng Junyong, Deputy General Manager of Juliang Technology, delivered a keynote speech titled "Where Should Simulation EDA Tools Go in the AI Era" and accepted interviews from multiple media outlets, including EEFocus. Founded in 2019, this domestic EDA company is striving to find its own position in the global EDA market dominated by giants.
Deng Junyong, Deputy General Manager of Juliang Technology
The AI Era: Opportunity or Trap?
AI is profoundly transforming the landscape of the EDA industry. From generating RTL code to automatically creating test environments, and further to invoking simulation, formal verification, and debugging tools, Agentic AI has begun to permeate the EDA workflow. International giants are actively deploying their strategies: Synopsys has launched a fully autonomous design verification agent, claiming to reduce RTL verification time by 50 times; Cadence, at Computex 2026, unveiled the industry's first fully autonomous virtual agent AI design engineer.
Facing this wave, Deng Junyong's attitude is quite prudent. "The role and efficiency of AI are indeed real," he stated in his speech, "but personally, for domestic EDA, AI might not be an opportunity."
In his view, the gap between domestic EDA and international giants mainly lies in the maturity of point tools. Deng Junyong said, "We say AI might bring us an opportunity to overtake on the curve, but we must proceed with caution and not put the cart before the horse."
He further elaborated on the logic behind this judgment: "Without the maturity of point tools, it is impossible for agents to mature. When we look at AI, if we focus too much on agents, and if EDA companies put too much energy into agents, it is somewhat neglecting the primary for the secondary. What we should make up for now is not agents, but rather making point tools mature and commercialized."
This perspective appears particularly pragmatic against the industry backdrop. The global EDA market has long been dominated by three major giants, with barriers reflected in multiple dimensions such as core algorithms, process adaptation, foundry certification, customer workflow integration, and engineer ecosystems. In the short term, domestic vendors find it difficult to comprehensively replace overseas leaders in full-flow tools for advanced nodes. The breakthrough of domestic EDA is more likely to evolve from single-point breakthroughs to "premiumization" and full-flow integration.
Regarding the relationship between AI and physics solvers, Deng Junyong's view is equally clear: "Currently, AI is still responsible for expanding the search space, while physics solvers remain irreplaceable for the time being. We must not neglect the primary for the secondary. There are some problems AI cannot solve; for some problems, it can improve efficiency and play a supplementary role, but it can never completely replace physics solvers."
Data Security: Local Fine-Tuning is a "More Viable Path"
In the process of AI empowering EDA, data security has always been an unavoidable topic.
"In the semiconductor industry, whether it is IC design or system development, including packaging and PCB, data sensitivity is absolutely the top priority," Deng Junyong frankly stated in an interview. He cited the experience of "moving to the cloud" during the pandemic as an example: "A major reason for people's distrust is data security. When it comes to chips, companies internally set up isolated networks, making it impossible for data to be leaked."
When asked whether customers prefer to fine-tune models with local private data or accept pre-trained general models provided by EDA vendors, Deng Junyong's answer was clear: "Most customers tend to accept the general foundation models provided by EDA vendors, and then fine-tune them with local private data, fully leveraging the enterprise's existing technical resources through RAG technology."
He explained the logic behind this: "If you share your data, there is another doubt—for example, if my data from Company A is trained and then can be used by Company B, this is absolutely unacceptable. Therefore, the training of this data cannot be shared."
Regarding Juliang Technology's own product roadmap, Deng Junyong stated that the company is currently mainly focusing on simulation and optimization tools: "The AI-related parts at Juliang have not yet utilized data training. Our main layout is in algorithm optimization, acceleration, software automation, and workflow adaptive optimization, ultimately reflecting AI's enhancement of accuracy and efficiency."
Customer Stickiness: Leveraging "Backup" to Displace "First Choice"
For domestic EDA vendors, breaking the customer stickiness to international giants' tools is the most realistic challenge.
"I have been doing design for many years. When faced with two tools in front of me, I still unhesitatingly choose the foreign ones without any room for negotiation," Deng Junyong candidly said. "If you understand this level, you actually know this is a long-term process."
However, he also sees a breakthrough point: "Most companies have choices. Some companies might plan ahead, but they won't truly lean towards you. What should we achieve? Technically, we must first be able to replace existing tools and achieve similar accuracy."
Deng Junyong believes that the biggest advantage of domestic EDA lies in service efficiency: "The biggest problem with foreign EDA is—for example, if my product is going to be released tomorrow and a simulation problem occurs today, if I give you a problem, you can't even give feedback in a week. This is the biggest issue. We have encountered this with relatively large customers, whereas we can do it differently: you might give me the problem today, and I can give you the results the next day."
"Doing simulation now is not like before," he explained. "In the past, doing simulation actually relied on personal experience, and you could judge whether something was problematic. But now, no one dares to make such judgments—for example, when I do Advanced Packaging or high-computing-power chips, because the cost is extremely high, the simulation step is indispensable."
The strategy adopted by Juliang Technology starts with being a "backup": "Many customers will choose to use me as a backup—I originally might buy 10 sets of tools from major vendors, and can allocate 1 or 2 sets to you. This actually doesn't affect their design. For most designers and simulation engineers, cross-validation with multiple sets of tools is the most reliable. This is an opportunity for us to slowly win customer trust and break this stickiness."
SIDesigner: From Point Tool to "Golden Accuracy"
SIDesigner is currently Juliang Technology's flagship product. According to the company, this platform uses a True-SPICE simulation kernel with Golden-level accuracy as its engine, building a one-stop SI/PI circuit simulation and sign-off platform.
Regarding the claim of "Golden accuracy," Deng Junyong stated that this is not self-proclaimed by the company: "'Golden' was not actually shouted out by us; it was verified by customers' test data after they tested it. Golden requires years of accumulation. It is not rigorous to form Golden in just one or two years—because it takes many years for everyone to default to recognizing you as Golden."
He elaborated in detail on the process of tool refinement: "After the basic algorithms are implemented, the next step might target a specific application field. For example, our current tools might target the high-speed design field, which requires polishing and accumulation through a large number of customer cases. Any EDA tool, without being polished through a large number of cases, cannot truly become a commercial tool, no matter how good your theoretical algorithms are."
In terms of application scenarios, SIDesigner covers mainstream high-speed interfaces on the market: "Including HBM, UCIe (used in Advanced Packaging), DDR, LPDDR5 (used in CPUs or mobile phones), SerDes, etc., covering high-speed interface scenarios in server chips, AI chips, CPUs, and Advanced Packaging," Deng Junyong stated.
Ecosystem and Future: The Path of Long-Termism
Regarding the M&A trends and market landscape in the EDA industry, Deng Junyong's views are quite pragmatic.
"Everyone wants to do full-flow, but whether they can achieve it is another matter. Will they be acquired, or continue to expand step by step? Actually, it might depend on your company—sometimes it's out of your control," he said. "If you want to expand slowly upwards, unless your existing products can sustain a company like yours, or can sufficiently support your gradual outward expansion in the future, otherwise it will be very difficult."
Regarding the future path of domestic EDA, Deng Junyong emphasized "long-termism": "This is a long-term process, and it cannot be done without accumulation." He believes the advantage of domestic EDA lies in "if you can get very close to foreign tools, ensure your product's simulation accuracy is close, and if you can subsequently engage in some in-depth cooperation with them, this is what many customers are willing to see."
At the end of the speech, Deng Junyong summarized the "where to go" of simulation EDA in one sentence: "We must base ourselves on reality and expand step by step, rather than drawing a very big pie first and then slowly filling the holes. This is the route to take in the EDA industry, or in doing domestic EDA."
This is perhaps exactly the survival strategy chosen by Juliang Technology—this domestic EDA company founded only seven years ago—in a market surrounded by giants.