Samsung Electronics' foundry business is extending its eight-year-old 8-nanometer process with another generation. The move appears aimed at making better use of existing, mature deep ultraviolet (DUV) production lines rather than relying on expensive extreme ultraviolet (EUV) lithography equipment. Its targets include automotive, robotics and edge AI chips that do not need the costliest 2nm technology.
According to industry sources on Oct. 20, Samsung Foundry will introduce 8LPU+, its second-generation 8LPU process platform, before the end of the year. Compared with 8LPU, the new platform is expected to improve performance by 12%, reduce power consumption by 15% and shrink logic area by 8%.
Samsung's 8nm technology took shape with the development of 8LPP in 2017. It represents the company's last DUV generation before the transition to EUV, using argon fluoride immersion lithography. Samsung unveiled 8LPU in 2018 with improvements in performance and density. The upcoming 8LPU+ is a further refinement of that platform.
Leading-edge processes such as 5nm, 4nm and 2nm offer greater computing performance but require EUV lithography for critical circuit layers. EUV machines cost hundreds of billions of won each, adding to wafer production costs. Supply is also constrained because ASML is the world's sole manufacturer.
By contrast, 8nm chips can be manufactured on existing DUV lines whose equipment has already undergone depreciation. Years of accumulated yield data make production costs more predictable, while existing automotive and industrial chip qualifications help facilitate supply to automakers and automotive electronics suppliers.
AI-enabled microcontrollers (MCUs) and automotive zone control units (ZCUs) are expected to be key targets for 8LPU+. Automakers are increasingly adopting zonal architectures, dividing a vehicle into physical zones with a controller assigned to each. Depending on the design, this can increase demand for controllers serving areas such as doors, the body and the powertrain.
MCUs and ZCUs used in vehicles and industrial robots must operate reliably in harsh environments. Operating temperature range, long-term supply availability, embedded memory and unit cost matter more than peak computing performance.
Reducing chip size is another priority. Even when the computing logic in an edge AI chip is made smaller, input/output (I/O) circuitry—which connects the chip to other components—and embedded nonvolatile memory (eNVM), which retains data without power, can still occupy substantial space. The 8LPU+ platform integrates 3.3V planar I/O devices into the 8LPU FinFET process, reducing I/O area by approximately 10%.
The platform also addresses standby leakage, the current that continues to flow when a vehicle is switched off or a robot is idle. It incorporates ultra-low-leakage SRAM (ULL-SRAM) and high-threshold-voltage devices (cHVT). Its embedded magnetoresistive RAM (eMRAM), which retains data without power, supports Automotive Grade 1 (AG1) requirements. The platform lists a bit-cell area of 0.017 μm² and memory density of approximately 20 Mb/mm², allowing more memory to fit into a given area and helping reduce controller chip size.
“Existing 8nm customers in Samsung Foundry's ecosystem, along with fabless chipmakers seeking automotive qualification, are likely candidates for the new 8LPU+ process,” an industry official said. “Design houses and application-specific integrated circuit (ASIC) companies in South Korea and overseas already have experience producing edge AI system-on-chips (SoCs) with 8nm eMRAM, which should make adoption easier.”
