Ultra-high-speed and low-power DDR5 is now available

As a veteran "lazy cancer" patient, you may have become accustomed to "talking but not doing" with the Samsung Galaxy S9丨S9+.

Hi, Bixby, help me identify this object.

In a blink of an eye, you know everything about this unfamiliar bicycle. The brand, features, price, and purchase information are all displayed on the mobile phone screen. This user experience not only changes the traditional information acquisition mode, but also has the characteristics of "viral" dissemination. As image recognition and machine learning become more and more popular, our knowledge structure and the way we understand the world will change accordingly.

When it comes to Samsung Bixby's image recognition function, whether it is a concrete object like a bicycle, a small and exquisite food like red wine, or even a landmark building, the Bixby built into the Samsung Galaxy S9丨S9+ can accurately identify it. Behind this is actually an extremely rigorous judgment process.

We know that the process of "seeing things" by machines and human eyes is different. The human eye will look at the overall picture first and then pay attention to the details, while machine image recognition will first scan the detailed features of the object. For example, the letter A is pointed, and the letter P has a bend, which is where the image contour has the largest curvature or the direction suddenly changes. These scattered features are then pieced together into complete information, and this discrimination mechanism is used to compare with the template features in the database to identify the object.

If a certain type of object cannot be recognized, or an error occurs, its image features can be relearned through correction and recorded, so that it can be successfully recognized next time. Therefore, the learning ability of Samsung Bixby should be better with use, just like the human brain.

Among all these, the response speed of the mobile phone is crucial. We all know that Samsung Bixby has its own independent dual-core engine, from voice recognition, to opening the camera and scanning the object, to local/network comparison, and finally returning the result, and even adding the learning and proofreading process, the whole process is almost in the blink of an eye, which is inseparable from the fast response of the DDR4 RAM carried by the Samsung Galaxy S9丨S9+. Because the role of RAM determines that it has the ability of short-term memory and fast access, which is crucial for image recognition and machine learning.

This is just a typical scenario of RAM and machine learning working together. In today's smartphone world where "experience is king", feedback processing for VR, AR, streaming video, online live broadcast, large-scale online games, and any other mobile phone needs to respond within milliseconds. A delay of half a second will result in being left behind by the opponent on the grounds of "poor experience", so the high-speed RAM built into smartphones, tablets, and car-mounted equipment has become the "main weapon" in this speed war.

DDR memory has gone through five generations of development in the past 21 years since its birth. To be fair, this 21-year history is the evolution of Samsung Electronics' DRAM manufacturing process and technology. Anyone who knows a little about this will know that Samsung Electronics has greatly promoted the development and evolution of the chip field, allowing DRAM products to double in speed and achieve breakthroughs in manufacturing processes in the past few years.

As the storage and processing hub, DRAM has the most basic and most difficult requirements of fast speed, low power consumption and large capacity. The development in the past 21 years has enabled the processing capabilities of mobile devices such as mobile phones to surpass traditional desktop devices, and has also made users more and more picky in this "speed race".

DDR 4, announced in 2007, has a 30nm process, a 1.2V operating voltage, a regular bus clock rate of 2133MT/s, and an "enthusiast" level of 3200MT/s. These gorgeous numbers seemed to completely surpass DDR3 at the time, but they were not realized until 2011 when Samsung Electronics completed the testing and manufacturing of DDR 4 DRAM.

The 11 years since DDR4 became popular in mobile phones, tablets, and smart car devices have also been the 11 years with the fastest development of mobile smart devices, and the improvements in DDR memory technology and speed have played an indispensable role in this.

Actually, I have foreshadowed it before. Why do we say that DDR has gone through 5 generations of development? This is all thanks to Samsung Electronics. On July 17, Samsung announced that it had successfully developed the industry's first LPDDR5-6400 memory chip, based on a 10nm-class (10-20nm) process. It is reported that the LPDDR5 memory chip has a single capacity of 8Gb (1GB), and a prototype of a module with an 8GB capacity has also been made and functional verification has been completed.

The industry's first DDR5 chip has monster-level performance: its memory speed (pin bandwidth) is up to 6400Mbps, 1.5 times that of LPDDR4X's 4266Mbps, and can transmit 51.2GB of data per second. At the same time, power consumption is reduced by up to 30% compared to LPDDR4X, mainly due to the addition of technologies such as dynamic voltage adjustment, avoidance of ineffective consumption, and deep sleep.

Are the numbers too boring? Let me explain the meaning behind these numbers one by one, and more importantly, how they will directly affect your mobile terminal experience.

First of all, a dazzling number is the data transmission rate of 51.2GB/s. What does this mean? According to the current high-definition video source standard of mainstream video websites (3.7GB per movie), 14 1080P movies can be transmitted in one second. You may question the significance of such a high speed? For mobile devices, the 5G network standard is imminent. The currently published 5G draft has a downlink rate of 20Gbps. It can be seen that under ideal network conditions, the processing capacity of DDR5 per second has more than 20 times the redundancy of the 5G transmission rate, and it is more handy in the face of the high bandwidth of 5G. Therefore, we can say that DDR5 is almost the standard configuration of mobile devices in the 5G era, otherwise it will become the short board of the barrel.

As for power consumption reduction, we know that with the effect of Moore's Law and the activation of the 10nm process, the power consumption of each generation of DDR will be reduced. However, the power consumption of DDR5 has been reduced by 30% compared to DDR4X, which is already a very significant improvement. For mobile devices such as mobile phones and tablets that attach great importance to standby battery life, the main means of energy saving is the change of screen material (such as activating OLED screen) and energy saving of onboard chips. Although we cannot quantify the role of DDR5 in improving the energy saving of the whole machine for the time being, the RAM alone can save 30% of power consumption, which is already very considerable.

Deep sleep technology is especially important here. We know that RAM’s role determines that it is an operating mechanism that requires fast response, but is also extremely fragmented in use. Moreover, the usage habits of smart devices are mostly “intensive and fragmented use during the day, and long standby at night without shutting down”. In this kind of peak-to-trough regular switching schedule, deep sleep like RAM is needed to maximize power consumption.

When it comes to power consumption, we have to mention the 10nm process that is closely related to it. The development of RAM in the past 21 years is the same as other chip components, which are based on the continuous improvement of the process. Compared with the previous generation 30nm technology, 10nm is much more compact in process structure. The direct benefit is the aforementioned reduction in energy consumption and the reduction in the overall size of the SoC. For mobile devices where every inch of space is valuable, the saved space will undoubtedly leave more room for designers to play. Previous comparisons have shown that the density of 10nm is 2.7 times higher than that of 14nm, not to mention the degree of density improvement compared to the previous generation 30nm.

At the same time, the launch of the 10nm process also demonstrates Samsung Electronics' technical strength in chip manufacturing. Although 10nm is a necessary step on the road to chip refinement, it is also the most difficult hurdle to overcome. Currently, only a few large manufacturers can ship stably using the 10nm process. Since the refinement of the process increases exponentially compared to the previous generation, the fault tolerance of the process, tools, quality control and other links also increases accordingly. Any particulate matter contamination may lead to a decrease in the yield rate. But what can be reassured is that Samsung Electronics' reserves in semiconductor processes are far more than this.