Samsung PM1743 is a solid-state drive designed specifically for enterprise servers, using the latest PCIe 5.0 interface and Samsung's advanced 6th generation V-NAND flash memory technology.

In terms of performance, PM1743 has excellent read speed and random read speed. Its sequential read speed is up to 13000MB/s and it can achieve 2500K input/output operations per second (IOPS). Compared with products based on PCIe 4.0, its speed is increased by 1.7 times and 1.9 times. In addition, the writing speed has also been significantly improved, with the continuous writing speed of 6600MB/s and the random writing speed of 250K IOPS, which have increased by 1.7 times and 1.9 times respectively.

PM1743 also has higher energy efficiency, reaching 608MB/s per watt, which is about 30% higher than the previous generation product. This will significantly reduce server and data center operating costs and help reduce carbon footprint.

This series of SSDs offers a variety of capacity options, ranging from 1.92TB to 15.36TB. It is available in the traditional 2.5-inch form factor, while also supporting the 3-inch EDSFF form factor. EDSFF is an SSD form factor designed for next-generation enterprise servers and data centers that is growing in popularity.

Overall, the Samsung PM1743 is a solid-state drive with excellent performance, high energy efficiency, and suitable for enterprise servers. Its outstanding performance and multiple capacity options make it ideal for handling large amounts of data and high-speed data transfer.

The new Samsung SSD 980 is the first retail-ready entry-level NVMe SSD from Samsung. Compared to previous SSDs from Samsung, it lacks any three-letter suffix at the end: it's not a PRO, nor an EVO, or even a QVO drive, because it doesn't fit into the established conventions for those product tiers. The SSD 980 slots into the product stacl at the bottom, but leaves a big enough gap for another, possibly better, model to come in later. Samsung is marketing the SSD 980 as a successor to the 970 EVO.
 
Samsung's First Push into Entry NVMe at Retail
 
The SSD 980 is an entry-level NVMe SSD, using TLC NAND with a DRAMless controller. This makes it a new class of product for Samsung in the retail market, one it has never produced before. This kind of drive, with TLC and without DRAM, is already prevalent in the market from other turn-key solution vendors, and this type of drive has been quite popular with OEMs: it allows them to advertise a NVMe SSD without paying the cost of a high-end drive.
 
 
 
The new Samsung SSD 980 will come in three capacities, 250 GB, 500 GB, and 1 TB. In this review we have the two highest capacities, featuring a Samsung custom controller and a variety of TLC NAND (explained more in detail below). As an NVMe drive rated for PCIe 3.0 x4, peak sequential read speeds are rated up to 3.5 GB/s for the 1 TB drive. The drives will be rated up to 0.3 drive writes per day (so 300 GB/day on the 1 TB model) for five years.

 

 
The Samsung SSD 980 is built around its "Pablo" controller. This isn't actually a new NVMe controller, but it has has been shipping in the Samsung Portable SSD T7 family for over a year. This controller has half as many NAND channels as the controllers used in Samsung's high-end NVMe drives (four rather than eight), and no DRAM interface. Using the high NAND IO speeds supported by recent generations of 3D NAND, the Pablo controller is able to saturate a PCIe 3.0 x4 host interface with out needing eight NAND channels.
 
 
 
Gallery: Samsung SSD 980


 
The NAND is the same 512Gbit 128-layer 3D TLC found on the 2TB 980 PRO. Lower capacities of the 980 PRO get the smaller 256Gbit 128L parts, but on the plain SSD 980 Samsung is using the larger capacity die across all three drive capacities. The smaller 980 PRO models need to use the smaller dies in order to make full use of the 8-channel controller, but the plain 980 with a 4-channel controller speed is less of a priority and the 512Gbit dies should be slightly cheaper.
 
(Note: This review initially stated that Samsung was using 256Gbit 92L TLC on our 500GB SSD 980 sample and speculated that they were doing so on the 250GB model as well. Samsung has since confirmed that it's the 512Gbit 128L on all capacities, meaning we misread a 0 as a D in the NAND package part number.)
 
Updates with the SSD 980
 
With the SSD 980, Samsung has shifted to a more aggressive SLC caching strategy, more than tripling the maximum size of the cache as compared with the 970 EVO (Plus). They made a similar shift with the 980 PRO. This is following a general industry trend toward larger SLC caches, most noticeable on QLC NVMe SSDs, which helps improve performance on many benchmarks and some real-world workloads. While good for performance numbers in short burst tests, it typically comes at the cost of performance when the cache fills up, or the drive fills up. For an entry-level drive that isn't intended for heavy workstation type workloads, this optimization toward peak performance makes a lot of sense.
 
Even though it's a more low-end product than any of their previous NVMe SSDs, Samsung is still giving the SSD 980 a five-year warranty with a 0.3 DWPD endurance rating, the same as all their recent EVO-tier drives.
 
A key part of Samsung plans to achieve good real-world performance with the SSD 980 is the NVMe Host Memory Buffer (HMB) feature, which we have covered in depth in the past. In brief: the DMA capabilities of PCIe allow an NVMe SSD to efficiently access the CPU's main memory/DRAM. The NVMe host memory buffer feature allows the operating system to give the SSD a small amount of DRAM to use for the SSD's own internal bookkeeping. Performance isn't as good as having DRAM on the SSD itself, and the amount of memory allocated is usually pretty small, default of around 64 MB but up to 1 GB, but it's enough to improve performance on a lot of real-world workloads.
 
Assuming the drive's Flash Translation Layer has a typical design, a HMB of 64 MB is enough to cache the logical to physical address mappings for about 64GB of storage. That's a very tiny amount of host RAM, but 64 GB of storage is a lot more than typical consumer workloads need to actively use at one time. The key part here is 'in use' - the HMB will only fill (and expand) as data is used in the drive for that session. So even though a top game might be 250 GB to install, it might only use 50 GB to play on a particular level.

 

The Samsung SSD 980 requests 64 MB of host ram for its HMB, but is willing to make use of as little as 16MB. These values are similar to other DRAMless NVMe SSDs we've tested. For this review, we've run the synthetic benchmarks with HMB enabled (the default on recent operating systems) and with HMB disabled to illustrate its impact.
 
The exception is the Mushkin Helix-L with the Silicon Motion SM2263XT controller: we didn't see any significant performance differences between HMB on and HMB off even on tests designed specifically to illustrate the effect of HMB, so it's pretty clear that drive's firmware never gets around to actually using the 64MB it asks for.
 
Samsung and Entry Level: The History
 
Historically, Samsung has distanced itself from making true entry-level drives for the retail SSD market, preferring to hold onto some degree of premium status and retain a more comfortable profit margin. This is a strategy that it's  vertical integration has supported very well.
 
Samsung has covered some of the low-end before however, with the QLC QVO SATA SSDs. But even here, Samsung has retained slightly premium prices, so the main appeal has been how affordable these drives are at the extremely large capacities. Samsung has not yet brought QLC NAND to their consumer NVMe product line, so the other obvious formula for a low-end drive is to go DRAMless while sticking with TLC NAND.
 
The break from Samsung's typical PRO/EVO/QVO split is not entirely without precedent: in the past there have been a few other occasions where Samsung has needed an extra product tier, resulting in the 750 EVO and SSD 850. Since then, Samsung has added the QVO tier of products, but the 980 doesn't qualify because it's using TLC NAND rather than QLC. Samsung also doesn't have room to decrement any part of the model number since there's no gap left between their 800-series SATA SSDs and 900-series NVMe. So we're left with the plain 980 while Samsung keeps a spot open for a better 980 EVO.
 
The Competition
 
We weren't able to review the WD Blue SN550 when it launched at the end of 2019, but we obtained a sample to include in this review because the SN550 is the most important competitor to the Samsung SSD 980. The earlier WD Blue SN500 significantly raised the bar for DRAMless SSDs, and the SN550 improves on it in almost every way. The Samsung SSD 980 is rated for much higher peak performance on sequential and random IO, but for more real-world workload the SN550 won't be so easy to beat. We also have results from the older Mushkin Helix-L, though with apparently inoperative HMB support it's nowhere near as competitive. Also in the low-end NVMe market segment we have a pair of QLC drives: the Corsair MP400 with the 8-channel Phison E12 controller, and the Intel SSD 670p with the new Silicon Motion SM2265 controller.
 
In the more mainstream TLC NVMe market segment, the drive to keep an eye on is the SK hynix Gold P31. Like the Samsung SSD 980 it uses a four-channel controller, but the P31 has the usual full amount of on-board DRAM which makes it a higher class of drive. If Samsung releases a 980 EVO in the near future, it will almost certainly go after the P31's position in the market, though Samsung may also try for a slight performance advantage by including PCIe gen4 support.
 
We also have results from several other recent Samsung SSDs in different market segments: the 870 EVO TLC SATA, 970 EVO Plus high-end PCIe Gen 3, and 980 PRO high-end PCIe Gen 4. The latter was re-tested with updated firmware a few weeks ago, but we didn't find any notable performance changes with our test suite.
 

Samsung Foundry has started mass production of chips using its 6LPP and 7LPP manufacturing processes at its new V1 fab. The new facility employs one of the industry’s first production lines built from the ground up for technologies that heavily use EUV tools.
 
EUVL is a key enabler for Samsung’s next-generation leading-edge fabrication processes as it allows to reduce usage of multipatterning, and increases yields of chips architected for smaller technologies. Samsung currently uses EUVL equipment for its 6LPP and 7LPP nodes, and will expand its usage in future nodes such as 5LPE, 4LPE, 3GAE, and 3GAP processes.
 

 
Samsung Foundry’s 7LPP and 6LPP process technologies are used to make advanced mobile SoCs that will be shipped in Q1. The contract manufacturer of semiconductors does not disclose names of its client(s), but it is noteworthy that its 6LPP technology is not a part of its MPW (multi project wafer) shuttle program, which may indicate that the process is only available for Samsung itself and/or select customers who do not use MPWs.
 




 
Samsung’s 6LPP is a evolution of its 7LPP node that offers slightly higher (~10%) transistor density and lower power, but is compatible with, and can reuse IP initially developed for 7LPP. Furthermore, 6LPP supports smart structures for designers eager to invest in all-new IP, as well as multi diffusion break feature. Meanwhile, 5LPE promises to bring further advantages when it comes to power, performance, when compared to 6LPP.
 
 
 
6LPP looks to be a limited offer node for select customers, whilst Samsung's focus in the future looks to be on its new 5LPE node.
 

 
Samsung’s V1 fab is located in Hwaseong, Korea, adjacent to its S3 line, where Samsung started initial ramp of chips made using its 7LPP technology. The company began construction of V1 in February 2018, and initiated test wafer production in the second half of 2019. Samsung continues to ramp up its V1 fab now and says that by the end of the year its total capacity for EUV-enabled 7LPP and below nodes will be three times higher than it was in 2019. Meanwhile, cumulative investments in the V1 line will hit $6 billion by the end of 2020.


 

Samsung has begun mass production of the industry’s first 16 GB LPDDR5 memory for upcoming smartphones, such as the Galaxy S20 Ultra 5G handsets. The new DRAM devices not only offer higher capacity, but they are also 30% faster than previous-generation LPDDR4X-4266 used for previous-generation cellphones.
 
Samsung’s latest 16 GB LPDDR5 package consists of eight 12 Gb chips and four 8 Gb chips featuring a 5500 MT/s data transfer rate that can provide bandwidth of up to 44 GB/s. The very complex 12-device assembly is called a ‘mixed mode package’ because it contains DRAM devices that are accessed differently in different ranks. Samsung does not publish voltages of the new multi-chip 16 GB DRAM assembly, but only claims that it delivers more than 20% energy savings when compared to an 8 GB LPDDR4X package. Meanwhile, we do know that LPDDR5 memory chips feature a variable voltage that is up to 1.1 V.
 

 
The 12 Gb LPDDR5 devices are made using Samsung’s 2nd Generation 10 nm-class fabrication technology (also known as 1y nm), but the company does not disclose which manufacturing process is used to build 8 Gb LPDDR5 devices (though we are probably talking about 1y nm here as well).
 
Typically, DRAM makers make mass production announcements after they ship the first batch of products to a client. Considering that Samsung’s Galaxy S20 Ultra features 16 GB of LPDDR5 memory, it is more than reasonable to assume that it uses the company’s latest 16 GB LPDDR5-5500 DRAM package.
 

 
 
In the second half of this year Samsung plans to start production of 16 Gb LPDDR5 devices using its 3rd Generation 10 nm-class fabrication process (aka 1z). These chips will support speeds up to 6400 MT/s, but to take advantage of them Samsung will have to develop an SoC that supports such a high data transfer rate. Meanwhile, capacities of these upcoming LPDDR5 assembles remain to be seen.


 

As part of the company's second quarter financial earnings call, Micron has revealed that it is about to start volume production of its 4th Generation 3D NAND memory devices. Based around the company's new replacement gate (RG) architecture, the memory manufacturer is gearing up to begin production in the current fiscal quarter (Q3'FY20), with commercial shipments set to begin in the fourth quarter. Overall, this will mark the start of a major technology transition for the manufacturer.
 
 
 
As previously detailed by Micron, the company’s 4th Gen 3D NAND features up to 128 active layers and uses replacement gate (RG) technology, which replaces the traditional floating gate technology that has been used by Intel and Micron for years. The switch is a substantial design change, and an important one going forward, as it's at the core of Micron's long-term technology plans. It also happens to be the company’s first flash memory technology in quite some time that has been designed solely by Micron, and not in conjunction with former partner Intel. Micron hopes that switching to gate replacement will enable it to reduce die sizes, lower costs, improve performance, and enable easier transition to next-generation nodes presumably with more active layers. 
 
Micron does not have plans to transit all of its products to its 4th Generation RG-based 3D NAND technology, and it has already warned its investors not to expect a meaningful company-wide cost-per-bit reduction this year as result of this technology transition. Nonetheless, it is tremendously important to kick off volume production as early as possible because learning how to produce replacement gate 3D NAND with decent yields is important for Micron’s subsequent generation 3D NAND that is projected to be deployed broadly in FY2021 (starts in late September, 2020).
 
Micron said that it plans to start shipments of its 128-layer replacement gate-based 3D NAND products in the fourth quarter of its FY2020, which means this summer. Meanwhile, Micron yet has to disclose which products it plans to build using this technology.
 
Sanjay Mehrotra, CEO and president of Micron, said the following:
 
In NAND, we made significant progress on our replacement gate, or RG, transition and expect to begin volume production in our current quarter, with revenue shipments to follow in our FQ4. We expect replacement gate production to be a meaningful portion of our total NAND supply by the end of this calendar year.
 

Capping off a busy week for fab-related news, South Korea authorities this week gave SK Hynix a green light to build a new, 120 trillion won ($106.35 billion) fab complex. The fab cluster will be primarily used to build DRAM for PCs, mobile devices, and servers, using process technologies that rely on extreme ultraviolet lithography (EUV). The first fab in the complex will go online in 2025.
 
 
 
The new cluster will house four huge semiconductor fabrication plants, which will be located on a 4.15 million square-meter site, reports The Korea Herald. The four fabs will have a planned capacity of around 800,000 wafer starts per month (WSPM), which will make the site one of the world's biggest semiconductor production hubs. Keeping in mind that we are dealing with EUV fabs, it is not surprising that a huge 200,000-WSPM plant with EUV tools will cost SK Hynix north of $25 billion. The fab cluster will be located near Yongin, South Korea, 50 kilometers south of Seoul, according to Yonhap news agency that cites the Ministry of Trade, Industry and Energy.
 
The new fabs will be used to make various types of DRAM using SK Hynix's upcoming production technologies that will use extreme ultraviolet (EUV) lithography. And with a start date still years away, we're likely looking at a fab that will be used to manufacture DDR5, LPDDR5X, and other future types of DRAM.
 
SK Hynix reportedly plans to start construction of the first fab in the Yongin cluster in the fourth quarter of 2021. Given the expected size of the massive building and the amount of time needed to folly load it with production equipment, SK Hynix expects this first fab to be completed in 2025.
 
It is necessary to note that just several years ago SK Hynix and Samsung used to build fabs that could produce both DRAM and NAND flash memory – or at least be converted with a minimal amount of effort. This is not the case today as DRAM production now heavily relies on lithography equipment, whereas 3D NAND production uses loads of etching tools, which is why the fabs for different types of memory have to be equipped completely differently.
 
The fab cluster in Yongin will be SK Hynix's second major DRAM site in South Korea after the company's primary DRAM hub near Icheon that houses its M10, M14, and M16 fabs. The M16 fab was completed in February and will be used for DRAM production using SK Hynix's EUV-based 1a process technology starting the second half of 2021.
 

SEMI, an organization representing chipmakers and producers of semiconductor production tools, published this week that sales of wafer processing equipment has surged to an all-time record of $71.19 billion for 2020. In the lead-up to the current chip crunch, equipment sales to South Korea and China noticeably spiked, with fabs in the former buying 61% more gear than in 2019, while China has risen to become the largest fab tool customer of all of the nations.
 
 
 
Overall, sales of fab equipment surged 19% from $59.75 billion in 2019 to $71.19 billion in 2020, according to SEMI. The substantial increase was driven by several factors. First and foremost, the world now consumes more chips than ever, and that consumption will only grow over time. Secondly, the competition between TSMC and Samsung Semiconductor (which has Foundry and Memory divisions) is escalating and both companies are spending more money on semiconductor equipment. Thirdly, next-generation lithography equipment (both DUV and EUV) is getting more expensive, so are other tools used in clean rooms. And finally, China is intensifying its domestic semiconductor efforts amid the trade war with the U.S.

 
 
Chinese companies increased their spending on wafer processing equipment by 39% year-over-year in 2020 to $18.72 billion, an all-time record for the country. Various companies, both domestic and foreign, are ramping up production of logic and memory chips in China, so the surge was something expected.
 
Taiwanese manufacturers bought semiconductor tools worth $17.15 billion last year, which was flat with 2019. Now that UMC (which is the world's third largest contract maker of chips) is focused on specialty and mature processes, it no longer has to buy leading-edge equipment. By contrast, TSMC's purchases of new tools offset declines at UMC, but on the country level shipments of semi tools were flat year-over-year. Meanwhile, this is going to change in 2021 as TSMC plans to radically increase its spending on new fabs up to $28 billion in 2020 and intends to invest $100 billion in new plants and R&D over the next three years.
 
South Korean companies increased their annual spending on semiconductor equipment to $16.08 billion last year, a whopping 61% year-over-year jump. Samsung Semiconductor, which has foundry services for logic, DRAM, and NAND flash memory, has been setting records with its CapEx budgets in the recent years. Its rival SK Hynix has also been increasing procurement of wafer processing equipment. As a result, in 2020 South Korean companies spent about the same amount of money on fab tools as Japan, North America, and Europe combined.
 
And though Japan is no longer a microelectronics mecca, but Japanese companies still spent $7.58 billion on fab tools last year, up 21% from 2019. A significant share of that expenditure likely belongs to Kioxia and Western Digital that constantly buy new equipment for their 3D NAND operations, and there are a number of other companies in Japan that produce more specialized semiconductors.
Meanwhile, tool purchases by American fabs actually dropped by 20% versus the previous year, sinking to $6.53 billion for 2020. The US is still the runaway leader for chip design, so the drop serves to widen the gap between how much is designed in the country versus how little is fabbed there. Overall it looks like the tables are going to turn in the coming years as Intel, Samsung Foundry, and TSMC begin to equip their new fabs in the USA; but for now, fab tool shipments are down significantly.
 
Finally, European fabs increased their purchases of new tools by 16% last year, totaling $2.64 billion invested in new equipment. As Intel brings its 7 nm fabrication process to Ireland in the coming quarters, the company will increase its spending in Europe, so fab tool sales there should see at least a temporary spike in the future.
 
In fact, tool sales are likely to spike everywhere for 2021 and beyond. While SEMI doesn't directly publish any outlooks for future sales, it's clear that the ongoing chip crunch has set the stage for a surge of additional equipment sales, as fabs are overwhelmed with orders despite already operating at full capacity. So, already fully booked for quarters to come, the need for new fab tools will only be increasing.
 

Capping off a busy week for fab-related news, South Korea authorities this week gave SK Hynix a green light to build a new, 120 trillion won ($106.35 billion) fab complex. The fab cluster will be primarily used to build DRAM for PCs, mobile devices, and servers, using process technologies that rely on extreme ultraviolet lithography (EUV). The first fab in the complex will go online in 2025.
 
 
 
The new cluster will house four huge semiconductor fabrication plants, which will be located on a 4.15 million square-meter site, reports The Korea Herald. The four fabs will have a planned capacity of around 800,000 wafer starts per month (WSPM), which will make the site one of the world's biggest semiconductor production hubs. Keeping in mind that we are dealing with EUV fabs, it is not surprising that a huge 200,000-WSPM plant with EUV tools will cost SK Hynix north of $25 billion. The fab cluster will be located near Yongin, South Korea, 50 kilometers south of Seoul, according to Yonhap news agency that cites the Ministry of Trade, Industry and Energy.
 
The new fabs will be used to make various types of DRAM using SK Hynix's upcoming production technologies that will use extreme ultraviolet (EUV) lithography. And with a start date still years away, we're likely looking at a fab that will be used to manufacture DDR5, LPDDR5X, and other future types of DRAM.
 
SK Hynix reportedly plans to start construction of the first fab in the Yongin cluster in the fourth quarter of 2021. Given the expected size of the massive building and the amount of time needed to folly load it with production equipment, SK Hynix expects this first fab to be completed in 2025.
 
It is necessary to note that just several years ago SK Hynix and Samsung used to build fabs that could produce both DRAM and NAND flash memory – or at least be converted with a minimal amount of effort. This is not the case today as DRAM production now heavily relies on lithography equipment, whereas 3D NAND production uses loads of etching tools, which is why the fabs for different types of memory have to be equipped completely differently.
 
The fab cluster in Yongin will be SK Hynix's second major DRAM site in South Korea after the company's primary DRAM hub near Icheon that houses its M10, M14, and M16 fabs. The M16 fab was completed in February and will be used for DRAM production using SK Hynix's EUV-based 1a process technology starting the second half of 2021.