Samsung Foundry to Unveil 8nm Gen-2 Node by Year-End

A machine that costs several hundred billion won pushes up the cost of every chip it helps make. Samsung Foundry has revived a way to make chips without that machine. The 8LPU+, a second generation of its 8nm process that is now more than eight years old, is set to be unveiled this year.

Samsung's 8nm effort took shape in 2017 with the development of “8LPP.” That process was the last DUV generation before EUV in Samsung's advanced-node lineup. DUV, or deep ultraviolet lithography, refers to the machines that use light to etch circuits onto a wafer. In 2018 came the 8LPU, with enhanced performance and density, and the 8LPU+ is the second generation of that line. Today's leading-edge nodes require EUV scanners for this key layer, and since ASML is the only company that makes them, supply is limited as well.

According to industry sources, the improvements over the previous generation amount to 12% higher performance, 15% lower power consumption and an 8% smaller logic area. The company also worked to shrink the chip. In edge AI chips, the input/output (I/O) circuits that exchange signals with the outside world and the embedded non-volatile memory (eNVM) that retains data even when power is off take up a large share of the area. By adding 3.3V planar I/O devices, the 8LPU+ cut I/O area by about 10%. Standby leakage, which drains current from parked cars or idle robots, was also addressed with ultra-low-leakage SRAM and high-threshold-voltage devices. The eMRAM, a magnetic memory that keeps data without power, is used at automotive grade 1 (AG1) level, with a bit-cell area of 0.017 µm² and a density of about 20Mb/mm².

The chip is aimed at automotive and industrial robotics. Recently, carmakers have begun adopting a “ZCU architecture” that divides vehicles into zones, each with its own controller, and under this design the number of controllers needed for doors, body and powertrain systems increases. Industry observers expect existing 8nm customers and fabless and design-house companies seeking automotive qualification to benefit first. The fact that domestic and foreign companies have already tested edge AI chips using 8nm eMRAM also lowers the barrier to adoption. However, this report did not confirm when mass production will begin, how much the chip will cost or which customers will use it.

All figures are industry-sourced and have not been verified by independent bodies. Even so, the reason this process is drawing attention is clear. It opens the way for chips that do not require the latest equipment to be used in greater numbers in every car.