Memory Market: More Than ASPs At Risk

By Adrienne Downey and Jim Feldhan

In June 2016, the memory market emerged from its slump after reversing its 12-month ASP decline. Since then, we’ve seen a strong rebound for ASPs in both DRAM and NAND. Contributing to this recovery was the increasing demand in memory content per device across all end markets combined with a more controlled capital investment over the past several years. The surge in memory revenue is driving the semiconductor industry to double digit growth rates, but there are risks.

Between 2012 and 2016, the industry’s capital expenditure for memory increased as a percent of total capital expenditures. Today, capital expenditures are driven by the need to add capacity for new process technologies, which have more process steps. The production of 3D NAND requires more equipment which has also resulted in an increase in capital expenditures. In 2012, only 30% of that year’s expenditures were devoted to memory. The memory portion has increased each year and by 2015 peaked at 39%, falling back to 36% in 2016. Semico expects the percentage of memory investment will go back up to 39% in 2017. (See the graph below).

Breaking out the capital expenditures by company reveals some interesting trends. Back in 2012, Samsung represented 56% of the memory capital expenditures. By 2016, Samsung only represented 37% of memory spending. SK Hynix was the second biggest spender in 2012 at 20.3% and dropped to third largest in 2016 at 21.6%. The Korean-based firms made up 76% of the total memory expenditures in 2012 and 58.8% in 2016. Between 2012 and 2016, Micron, Toshiba, Intel and Western Digital (SanDisk) increased their spending. Also during that time frame, Intel logged the fastest growth in memory capex at a CAGR of 29%, mostly due to its small starting base. Intel’s spending is expected to increase in 2017 as the company focuses on 3D NAND and XPoint memory, particularly for supporting computing performance which will drive MPU sales.

Micron’s expenditure growth rate was the second fastest at a CAGR of 28.7% as they expanded and upgraded wafer fabs. In 2016, Micron was the second largest spender, and in 2017 they will drop back to third behind SK Hynix again. SK Hynix has experienced a 9.4% CAGR for its capital expenditures. As a percent of the total, SK Hynix has hovered between 19% and 24% with 2017 forecast to be 21.4%.

Between 2012-2016, Toshiba and WD (SanDisk) both increased their spending by 13.7% and 12.3% on a CAGR basis. The table below depicts the narrative described above.

Of course, we can’t forget the investment going on in China. There are several DRAM and NAND fabs which are scheduled to ramp production in late 2018 or 2019. New players always have a way of disrupting the status quo.

Increases in average selling prices for memory products are the main factor driving the recent memory revenue growth. Memory prices hit a low in June of 2016, and by the end of the year prices for DRAM had increased 30.8%. Likewise, the NAND market saw prices rise 44.2% by year-end. Prices continue to rise in the first half of 2017. Just this year, DRAM ASPs have increased by 47.2% since December 2016. NAND ASP growth has moderated somewhat but has still increased 20.6% since December 2016.

In addition to conservative wafer fab expansion and the growing demand for memory content, there are a few manufacturing trends that have contributed to the increased capacity utilization. Migrating to finer geometries and higher density DRAM designs have required more process steps, more metal layers and multi-patterning. As a result, wafers are in the fab longer, which results in fewer wafer outs per month. In 2017, the 8Gb device will be the largest volume runner with the lowest cost per bit among the DRAM product families.

NAND manufacturers are continuing to transition to 3D NAND. The implementation of 3D NAND has been a boon to equipment vendors. 3D NAND structures require significantly more etch and deposition steps typically resulting in more time in the fab. From a bit standpoint, 3D NAND fabs can produce a comparable amount, or even more bits than a planar fab.

While technology continues to improve and fab capacity continues to expand, some level of risk will always loom over the industry, whether from earthquakes or political tensions. Due to recent geo–political instabilities, Semico took another look at how much capacity is located in South Korea. Using Semico’s Fab Database, Semico can plot the location of all the fabs around the world. How much of the world’s memory capacity is located in South Korea?

Both Samsung and Hynix operate wafer fabs in South Korea. The map below shows the location of most semiconductor fabs in South Korea, highlighting the memory fabs belonging to Samsung and Hynix. The S = Samsung, H = SK Hynix and the number equals the number of fabs at that location (S-1 means one Samsung fab at that location). Samsung has 8 fabs and 5 planned fabs in Hwaseong for memory, although one of those is brand new—S3/17 Phase 2. There is also a brand-new monster-sized Samsung 3D NAND fab in Pyeongtaek. SK Hynix has 3 operational fabs and 1 planned fab in Icheon. SK Hynix also has 3 operational and 1 planned fab in Cheongju. Total memory wafer capacity for Korea is about two-thirds of the world’s total memory production.

All of these fabs are in the range of North Korea’s old artillery or relatively new 300 millimeter rocket launchers that can simultaneously fire 12 rockets with high explosive and incendiary or chemical warheads.

While an attack from North Korea ranks on the low side of the probability scale, the current situation is very tense. It’s also interesting to note that the US military relocated the US 8^th Army from its base in Seoul to Camp Humphreys in Pyeongtaek, farther south of Seoul.

Adrienne Downey is the Director of Technology Research, Manufacturing at Semico Research.