FT article update says it has been done
Chinese semiconductor thread II
- Thread starter vincent
- Start date
Working together with Feiteng and ZTE, Sinovation uses AI to drive verification innovation and directly address the "efficiency pain" of chip verification
On April 16, as a leading system-level verification EDA solution provider in China, Xinhuazhang worked with Feiteng Information Technology and ZTE Microelectronics to give joint speeches at DVCon China, the most influential conference in the field of IC design verification. They used actual cases to explain how "AI+EDA" can reshape verification efficiency by targeting the "hard bones" in verification in various scenarios, so that everyone can see the solid implementation of domestic verification EDA technology and the power of ecological collaborative innovation.
Solving the Dilemma of Complex Operator Verification from "High Threshold" to "High Efficiency"
In the design of high-performance CPUs and AI chips, the verification of complex operators such as floating-point operations and matrix multiplication has long faced the challenges of "incomplete coverage and low efficiency". Traditional simulation methods are difficult to exhaustively enumerate massive boundary conditions, and the high threshold of manual formal verification has daunted many teams. Xinhuazhang and Feiteng jointly tackled the problem and deeply integrated AI algorithms into Xinhuazhang's GalaxEC-HEC high-order equivalence verification tool. Through intelligent case splitting and mathematical proof engines, accurate mapping from algorithm level to RTL level was achieved - the proof time of double-precision multiplication and addition operators was reduced from 89 hours to 11 hours, and single-precision multiplication and addition was compressed from 86 hours to 1 hour.
LLM makes complex assertion generation move from "manual journey" to "intelligent generation"
SystemVerilog assertions (SVA) are crucial to ensuring the correctness of the design in formal property verification (FPV), but the manual writing of complex sequential logic is time-consuming and labor-intensive, and it is easy to cause verification vulnerabilities due to missing boundary conditions. Aiming at this pain point, Xinhuazhang and ZTE Microelectronics jointly developed SVA generation based on large language models (LLM) and introduced the industrial-grade innovative evaluation system SVAEval. This framework significantly improves the quality of SVA generated by large language models through iterative prompt optimization and indicator evaluation. In real project verification, the system demonstrated strong scenario adaptation capabilities: in the face of complex requirements such as "target signal matching within 10 cycles after signal change" and "asynchronous clock domain assertion synchronization", it can automatically generate assertion code containing temporary variables, time windows, and cross-clock domain processing, covering more than 20% of boundary scenarios that are easily missed in traditional manual writing.
Dude I don't think Trump even know what EDA means.
"Stooges of the Department of Commerce orders chips designers to stop selling to China"
Brite Semiconductor Launches 28HKC+ Process Platform TCAM IP
Canxin Semiconductor (Shanghai) Co., Ltd. (Canxin Shares, 688691 ), a one-stop customized chip and IP supplier, announced the launch of Ternary Content - Addressable Memory (TCAM) IP based on the 28HKC+ 0.9V/ 1.8 V platform . This IP has high frequency and low power consumption characteristics . As the number of scenarios that require efficient search, such as fast processing of routing tables and access control lists ( ACLs ) in network equipment continues to increase, this IP will be widely adopted by high-end switches, routers and other chips.
The TCAM IP released by Brite Semiconductor this time includes a fully self-developed Bit cell , which complies with the logic design rules and has a high yield rate. In terms of reliability, this TCAM is highly resistant to process deviations and has the characteristics of high reliability. In terms of power consumption, the match line segmentation control is adopted , which can save about 30% of power consumption. In terms of speed, it can support up to 800Mhz and has a fast search speed. In addition, customers can compile TCAM cells of different sizes according to their needs, which has the characteristics of highly flexible design.
At present, the IP has been successfully taped out and has passed the chip function test and performance test of the 28HKC+ process platform , and all indicators have reached the expected standards . With the continuous improvement of the demand for high-speed networks and various intelligent applications , the importance of TCAM IP in these applications has become increasingly prominent. Brite Semiconductor can provide complete TCAM IP and customized chip solutions to help customers with product design.
Can China pirate the EDA software and continue to use it? Softwares are easy to pirate.
this is for you..
Huawei has complete EDA stack fully functional at least since 2023. officially announced by Chairman itself. its now upgraded to 7nm and below.
But Is Huawei selling that software? Or is it just internal tool. There are many Chinese chip design companies that need EDA.
Huawei said back then.. its fully operational.
China has now Four different companies in EDA sector, covering wide range of integrated circuit.
Empyrean, Xpeedic, X-EPIC, Primarius .. there is enough details on this thread.
back in 2023, According to Empyrean, its analog tools can already partially support 5nm process, while its digital tools can fully support 7nm process
Software literally has no problem. let USA block the Software sales.