SK hynix broke new technological ground to achieve the recent landmark development of the world’s first DDR5¹ product built using the 1c² node. Boasting improved operating speed and power efficiency compared to the previous generation, the 16 gigabit (Gb) 1c DDR5 represents a monumental leap forward in DRAM process technology.

¹Double Data Rate 5 (DDR5): A server DRAM that effectively handles the increasing demands of larger and more complex data workloads by offering enhanced bandwidth and power efficiency compared to the previous generation, DDR4.
²1c: The sixth generation of the 10 nm DRAM process technology, which was developed in the order of 1x-1y-1z-1a-1b-1c.

The groundbreaking achievement is just the latest in a long line of breakthroughs by the company to advance its DDR5 lineup. This remarkable progress is not only a testament to SK hynix’s technological prowess but also to its innovative approach to the validation process.

This episode of Rulebreakers’ Revolutions will focus on how the company’s differentiated validation strategy is enabling it to navigate the challenges posed by a diversifying server CPU market, contributing to SK hynix’s DRAM leadership including its cutting-edge DDR5 server DRAM.

The Mission: Strengthening Validation in an Increasingly Diverse CPU Market

SK hynix has faced various challenges throughout the successful development of its latest 1bnm and 1cnm DDR5 products. As each generation of DDR5 is based on a new DRAM process technology, a rigorous validation process is required to ensure the products’ performance and reliability as well as their compatibility with customer systems. This means that the company has had to continually adapt and strengthen its approach to validation for next-generation technologies.

In addition to validating new technologies, SK hynix also had to respond to the growing diversity of server CPU suppliers in the market. Traditionally, Intel has dominated the sector which has resulted in semiconductor companies primarily focusing their efforts on validating their products with the U.S. tech giant. However, while Intel still leads the sector, AMD and Arm-based suppliers are gradually increasing their market share, especially in cloud and specialized workloads, creating a more fragmented server CPU landscape.

Amid these shifting market dynamics, SK hynix is required to ensure its DDR5 products’ compatibility and reliability across a broader range of server CPU architectures. This is because CPU companies integrate their chips into a wide range of hardware, placing greater pressure to ensure compatibility with various server CPU types, which is essential for DDR5’s widespread adoption. In particular, early alignment with customers during the product planning stage is increasingly important to meet the needs of various companies and handle numerous server CPU types.

Faced with the challenge of validating new technologies in a diversifying server CPU market, SK hynix is consistently refining its validation methods to ultimately solidify its leadership in the server DRAM field.

SK hynix employs a differentiated validation approach to ensure DDR5’s comparability with various server CPUs

Compatibility, Collaboration, & Customized Testing: Three Cs Key to Optimized Validation

Although SK hynix had an established validation procedure for its products, the company has adjusted its process for the latest DDR5 products. Typically, validation begins in the pre-development stage by verifying that the design meets the specifications required by server customers and adheres to JEDEC³ standards. The company then prepares memory validation samples in collaboration with external partners and aligns the test environment with SoC⁴ companies, which play a key role in validation. While CPU companies are the customers for DDR5, SoC companies provide reputable third-party validation to verify the product’s readiness for real-world system application. Continuing with the validation process, the next stage involves internal testing to identify and resolve any defects. The samples are then sent to SoC companies for further tests to complete the process.

³JEDEC Solid State Technology Association: With over 350 member companies, JEDEC is the global leader in developing open standards for the microelectronics industry.
⁴System-on-chip (Soc): An integrated circuit that combines all the components of an electronic device onto a single chip.

For its recent DDR5 products, however, the company is taking a rulebreaking approach to validation that sets it apart from the rest of the field. This unique method is highlighted in the ongoing validation process for the 1bnm and 1cnm DDR5, which features several differentiated strategies.

For example, to ensure compatibility with a wide range of server CPUs, SK hynix is conducting validation on a wide range of systems—even those that have yet to be released. This involves close collaboration with SoC companies to discuss required technologies and perform co-validation, ensuring that potential issues with samples are addressed early in the development process.

As well as working with external SoC companies, SK hynix is also conducting improved internal collaboration throughout the validation process to enhance the completeness of the 1c DDR5. Departments responsible for process, design, and testing are closely working together to ensure the cost efficiency of test infrastructure and optimize the sample management and testing process. Moreover, the departments are identifying and improving potential defects in advance through rigorous simulation and aging tests, thereby securing the product’s reliability and stability.

Another key step in the validation process involves the development of customer-specific tests based on tailored validation scenarios. Recognizing that each customer has different requirements and various products, SK hynix evaluates and verifies various scenarios at the scale and volume testing stages. By predicting potential defects under actual usage conditions, the company aims to ensure that the 1c DDR5 performs reliably across a wide range of applications and platforms.

SK hynix’s validation strategy involves ensuring broad compatibility, internal collaboration, and customized testing

Validation: The Final Piece of the DDR5 Evolution Puzzle

Since SK hynix launched the world’s first DDR5 DRAM in 2020, rapid and reliable validation has played a crucial role in the company’s outstanding progress in the field. The company’s numerous milestones in DDR5 and DRAM scaling technology include the industry-first validation of its 1anm DDR5 with the 4th Gen Intel® Xeon® Scalable processor in January 2023. In May of the same year, the company announced it had developed the industry’s most advanced 1bnm DDR5 and begun validation with Intel.

While the semiconductor industry has faced increasing difficulty in advancing the 10 nm process technology, SK hynix overcame the obstacles thanks in part to its robust validation strategy. This is set to continue with the validation of its 1c DDR5 product, which aims to verify key specifications of the new DRAM.

The 1c DDR5 offers superior operating speeds and power efficiency compared to the previous generation

In comparison to 1b DDR5, the 1c DDR5 product offers 11% faster operating speeds of 8 gigabits (Gbps) per second and a more than 9% improvement in power efficiency. As part of the validation process which is progressing smoothly, SK hynix is currently working with server CPU suppliers to verify the product’s stable performance and ensure it meets the expected operational standards.

Looking ahead, the successful development of 1c DDR5 has established a benchmark for subsequent DRAM product lines to be developed with the 1c node, including HBM⁵, LPDDR⁶, and GDDR⁷. In terms of validation, the company is striving to enhance the efficiency and overall process for future products. In particular, the company aims to reduce risk factors in future validation processes, ensuring higher quality and reliability for its next-generation products.

⁵High Bandwidth Memory (HBM): A high-value, high-performance product that possesses much higher data processing speeds compared to existing DRAMs by vertically connecting multiple DRAMs with through-silicon via (TSV).
⁶Low Power Double Data Rate (LPDDR): A line of low-power DRAM for mobile devices, including smartphones and tablets, aimed at minimizing power consumption and featuring low voltage operation.
⁷Graphics DDR (GDDR): A standard specification of graphics DRAM defined by the Joint Electron Device Engineering Council (JEDEC) and specialized for processing graphics more quickly. It is now one of the most popular memory chips for AI and big data applications.

Rulebreaker Interview: Yoosung Lee, DRAM Server Product Planning

To find out more about the company’s innovative approach to validation, the SK hynix Newsroom spoke with Technical Leader (TL) Yoosung Lee of DRAM Server Product Planning. Lee’s department plays a key role in collaborating with customers such as SoC companies to validate products. He discussed how the company encourages employees to take a different approach to their work and the improvement plans for validation.

<Other articles from this series>

[Rulebreakers’ Revolutions] How MR-MUF’s Heat Control Breakthrough Elevated HBM to New Heights

[Rulebreakers’ Revolutions] How SK hynix Broke Barriers in Mobile DRAM Scaling With World-First HKMG Application

[Rulebreakers’ Revolutions] Innovative Design Scheme Helps HBM3E Reach New Heights

The post [Rulebreakers’ Revolutions] How SK hynix’s Server DRAM Validation Process Succeeds in a Diverse Server CPU Market first appeared on SK hynix Newsroom.

But how familiar people are about the history and rise of semiconductors is a different matter. To provide more context on these indispensable materials that power everything from household appliances to mobile phones, this series will trace back the origins of semiconductors and explain why they became such a significant part of everyday life as we know it.

Starting with ‘Computers and Transistors,’ a total of six chapters including ‘Process and Oxidation,’ ‘Photolithography,’ ’Etching,’ ‘Deposition,’ and ‘Metal Wiring’ will help explain the nature and processes of semiconductors. This series will put a special focus on the correlations between all of these technologies.

The Advent of Computers

As people continuously look for ways to simplify their day-to-day activities at home, work, and wherever they need to go, the need for technological devices has always been on the minds of innovative thinkers. Starting with simple machines that only knew how to make basic calculations, people eventually progressed to developing more advanced and accurate machines that would become of more practical use.

Inventing such a machine required tremendous contributions from various people. A major experiment that came out from one of these individuals was Charles Babbage’s Analytical Engine in 1871. Users could insert a thin plate called a punched card into the machine and perform a numerical calculation. When inserted into the machine, the inner analysis engine repeats various arithmetic operations according to specific commands and a result value is printed out from another part of the machine.

Although the Analytical Engine was never manufactured, it stands as an interesting case study. First, the Analytical Engine has all the elements of a computer. The punched card and the part of the engine where the result value is printed out is the same concept as the computer memory. So, the Analytical Engine is essentially the primitive CPU^* (Central Processing Unit).

^*CPU: Abbreviation for Central Processing Unit. A device that acts as the computer’s brain.

Simply put, the Analytical Engine was a computer that operated on steam and consisted of a memory and CPU unit made from pieces of metal and wood. Consequently, we can assume that people in the past knew how computers structurally functioned. It’s also clear that computers and ‘electronic circuits’ are completely different concepts. Therefore, there’s a need to know why electronic circuits became the heart of the modern computer.

Electrically Controlled Computers

Electronic circuits have advantages over other devices based on steam, manpower, or hydraulic power, because the control of signals is fast and efficient. Looking at steam, reaction rates are slow, as steam needs to physically reach a specific location. Furthermore, since steam is transmitted at a high pressure, the pipes need to be thick and, overall, it lacks efficiency. Now, let’s suppose there’s a device that opens and closes its door automatically when a rope is pulled. If steam is used as the energy source here, the operator will have to open the boiler valve and wait for the high-pressure steam to push the door in order for it to be closed. But when electricity is the energy source, a button and a motor is all that is required. The size of the entire device becomes smaller, while energy efficiency and reaction speed both increase.

▲ Figure 1. A steam-based automatic door (left) and an electric automatic door (right)

With the invention of electricity, controlling computers with it became the general trend. After numerous attempts to create an electricity-based computer, the ENIAC (Electronic Numerical Integrator And Computer) was eventually made. Unlike the Analytical Engine that used gears and steam, the ENIAC operated through the combination of a type of light bulb called the vacuum tube and various electronic circuits. By looking at the components of the ENIAC that resemble light bulbs, it’s quite clear that its energy source was electricity.

▲ Figure 2. ENIAC (Source: View Original Document)

The ENIAC was a huge computer that almost took up a whole room and used up to 170 kW of electricity, equivalent to operating 170 microwave ovens. Nevertheless, it was able to accomplish numerous tasks that were needed back then. Its operation speed was still a lot faster as it used more than 170,000 vacuum tubes instead of gears that squeaked and moved slowly. Since its development, the ENIAC contributed towards achieving many milestones, including formulation of the simulation methodology.

However, we know that the ENIAC’s performance was not even a match for the portable calculators of the 1990s. Efficiency was a major issue, and it wasn’t possible to supply these commodities on a large scale due to their size. This is why the world needed another innovation called the transistor.

The Emergence of Transistors

As aforementioned, the ENIAC was built using a vacuum tube similar to a light bulb. But it’s important to know why these devices were needed in the first place. People knew that the ability to control signals would lead to the creation of some type of computing device, such as the automatic steam door that we looked at above.

Thus, a computer is basically a device that adds a lot of inputs and outputs to an automatic steam door along with various logical structures that are added by connecting thousands of pipes in the interior. Automatic steam doors can only do simple tasks such as opening and closing the door. But a computer can be built when it’s possible to carry out more complex tasks such as simultaneously opening two doors with a single rope or making a safety door that does not close when a person is under the door. Ropes and steam pipes basically play the role of basic devices corresponding to a vacuum tube.

▲ Figure 3. An automatic steam door that opens multiple doors with one operation (left),
and an automatic door that opens only when two operators agree to open the door (right)

So, how can the performance of a steam computer be enhanced while adding extra functions to it? The number of steam tubes could be increased to add more functions, or a boiler with a higher pressure and temperature could be installed to increase the speed at which the steam rises. The problem with these solutions is that they are not easy.

Steam engines are very large by themselves, so adding a tube from the boiler to another area creates an even bigger burden on space. It requires too much energy, and the risks of explosion or other malfunctions also increase when trying to enhance the performance of the boiler. Vacuum tubes were merely the best devices available to engineers at the time. As they operate on electricity, there are no risks of explosions like with a high-pressure boiler. And the operating speed was clearly faster than the steam engine. Of course, there were frequent accidents such as individual vacuum tubes malfunctioning due to too much power being used. To make a better computer, it was necessary to find more advanced components.

In 1947, the transistor was invented. Transistors were innovative devices that could regulate the flow of large currents with very small currents. Scientists found that by using two types of semiconductor materials, as shown below, it was quite easy to disconnect and connect signals. Although the structure may look complex, the nature of its operation is essentially the same as controlling the movement of steam by pulling a rope. In the same year that the first transistor was invented, the BJT^* (Bipolar Junction Transistor), which is widely used to this day, was also invented. At this juncture, semiconductors also began to be known by the public.

^*BJT: The Bipolar Junction Transistor, within a semiconductor is a transistor made by using a PN junction, or the boundary between two domains of a P-type semiconductor and an N-type semiconductor.

▲ Figure 4. The structure of the transistor. Both N-type and P-type semiconductors are used.
(Source of the right image: The Understanding of the Semiconductor Manufacturing Technology, p. 143, Table 4-6)

Semiconductors for everyone: MOSFET’s Revolution and Its Manufacturing Technology

While working at Bell Labs in 1959, Dr. Martin Mohammed John Atalla and Dr. Dawon David Kahng developed a new type of transistor called the MOSFET (Metal Oxide Semiconductor Field Effect Transistor). The two scientists formed two types of semiconductor layers on a silicon disk and then placed metal on top of it to create a flat transistor. Although the operating principle of this transistor was slightly different, its usage was not too distinct from the transistors introduced above. But what made this transistor stand out was its productivity.

▲ Figure 5. Dr. Dawon Kahng’s MOSFET model structure (Source: Hanol Publishing Co., Ltd.)

Due to their flatness, numerous MOSFETs could be made on a silicon wafer simultaneously. If the outline could be made smaller, it was possible to make ten times more MOSFETs on wafers of the same size. Additionally, a set of already-connected MOSFETs could be manufactured simultaneously as well. Let’s say a CPU needs to be built using a BJT. No matter how efficient the BJT’s production process is, it’s necessary to solder hundreds of millions of BJTs together and attach them to the circuit board, as CPUs were made by connecting BJTs. As for MOSFETs, hundreds of millions of transistors are already soldered to the circuit board when they are produced.

Eventually, the whole point of semiconductor factories is to make MOSFETs more affordable. The succeeding chapters will explain how terms like exposure, etching, deposition, and other processes of making semiconductors contributed to producing affordable MOSFETs.

Read articles from the Front-End Process series

Read articles from the Back-End Process series

The post Semiconductor Front-End Process Episode 1: The Birth of Computers, Transistors, and Semiconductors first appeared on SK hynix Newsroom.