Abstract:
Samsung’s wafer foundry business will continue the 8-nanometer process, which has been around for more than eight years, for another generation. External interpretations believe that this move is intended to efficiently revitalize the existing mature deep ultraviolet (DUV) production lines and avoid the expensive extreme ultraviolet (EUV) lithography machines. This process is aimed at automotive, robotic, and edge AI chips that do not require the high cost of 2 nanometers.
According to industry news on the 20th, Samsung wafer foundry will launch the 8nm second-generation 8LPU+ process platform within the year. Compared with the previous generation 8LPU, the performance is improved by 12%, the power consumption is reduced by 15%, and the logic area is reduced by about 8%.
Samsung’s 8-nanometer process has been officially commercialized since the successful development of 8LPP in 2017. This process is also the last generation of DUV (argon fluorine immersion lithography) process in Samsung’s advanced process before the introduction of EUV. In 2018, Samsung launched 8LPU, which focuses on performance and integration density optimization; the newly released 8LPU+ this time is a second-generation version further improved on the basis of 8LPU.
Currently, cutting-edge processes such as 5 nanometers, 4 nanometers, and 2 nanometers have stronger computing power, but EUV lithography machines must be used for key lithography layers. The cost of a single EUV lithography machine is as high as hundreds of billions of won, pushing up wafer costs; and only ASML can supply it in the world, so equipment supply is limited.
In contrast, the 8-nanometer process can be produced on DUV equipment production lines that have completed depreciation and amortization. After long-term accumulation, the yield data is sufficient, the cost can be predicted, and it has obtained relevant certifications for automotive and industrial chips, which lowers the supply threshold for OEMs and automotive parts suppliers.
The core target products of 8LPU+ are expected to be AI microcontrollers (MCU) and automotive area controllers (ZCU). Recently, the automotive industry has generally adopted the ZCU architecture: the vehicle is divided into multiple areas, and each area is configured with a separate controller. After adopting this architecture, controllers need to be arranged in various partitions such as the door, body, and powertrain, and the demand for chips increases accordingly.
MCUs and ZCUs used in vehicles and industrial robots need to work stably in harsh environments. Compared with peak computing power, the operating temperature range, long-term continuous supply capability, built-in storage, and unit price are more core indicators.
Chip miniaturization is also a key point. Even if the computing core area of edge AI chips is reduced, the I/O responsible for external connections and the embedded non-volatile memory (eNVM) that retains data after power outage will still occupy a large amount of area. The 8LPU+ platform integrates 3.3V planar I/O devices on the 8LPU FinFET wafer, successfully reducing the I/O area by about 10%.
This process also improves the standby leakage current problem when the vehicle is turned off and the robot is in standby state, and is equipped with ultra-low leakage SRAM (ULL-SRAM) and high threshold voltage devices (cHVT). The magnetic storage eMRAM that saves stored data when power is turned off reaches automotive grade 1 (AG1) specifications; the unit area is 0.017μm² and the storage density is about 20Mb/mm². It can accommodate more storage in the same area, which is conducive to further reducing the size of the controller chip.
Industry insiders said: "Existing 8nm old customers in Samsung's wafer foundry ecosystem and fabless chip companies that want to obtain automotive certification will all be potential customers of 8LPU+. Domestic and overseas design companies and ASIC manufacturers already have experience in edge AI system-on-chip (SoC) based on 8nm eMRAM tape-out, and the entry threshold is low."
Comments