Site Search

Camera? Computer? How New Home Security Vision System Architecture Affects System Memory Technology Choices

Introduction

Artificial intelligence (AI) has been attracting attention for many years. AI) and machine learning (ML) technology-based products have begun to hit the consumer market.

According to the research and development team's announcement, it is said that it is difficult or impossible for machines to learn human-like skills and judgment in some applications such as autonomous driving. AI Despite the hype in some areas, it is also true that many products with machine learning capabilities are beginning to gain widespread interest among consumers. For example, intelligent vision-based security and home monitoring systems hold great promise. of a research company Strategy Analytics The company 2019 from the year 2023 During the year, the market value of the home security camera market increased to 80 from $100 million 130 to billion dollars 50 expected to grow by more than 10%.

One of the most suitable functions for machine learning technology is image recognition and scene recognition, and the development of intelligent cameras is expected. Using intelligence in your home vision system will allow you to:

・Detects the possibility that elderly people and children have fallen and injured themselves
Monitor the infant for normal sleep breathing
・Recognize your face to open the lock, or recognize your pet cat to open the cat door flap
・Detects suspicious movement outside the house and issues an intruder alarm

advanced image processing (Image Signal Processor / ISP), these new intelligent home vision systems are function-specific computers. The latest products in this field include: ISP Faster application code running in DRAM It uses a computer-like architecture that is stored in system memory and operates with low latency and high response.

This article explores how the new category of home vision systems DRAM verifying which technology to choose, ISP For the requirements of the base architecture, DRAM Here's how the manufacturer handles it.

System memory for full-fledged computer cameras

Modern home monitoring intelligent cameras use neural networks for image and scene recognition. These vision systems contain an inference engine. It is a software algorithm that "learns" to recognize types of images and scenes by comparing and analyzing thousands to millions of labeled images in a training dataset.

Figure 1: An example of an early AI-based home security camera, the NetGear Arlo. (Image credit: Scott Lewis under Creative Commons license)

Running an inference engine is a computationally intensive task, so some modern home vision systems employ high-performance ISPs incorporating AI propulsion engines from chipset manufacturers such as Ambarella. These ISPs typically feature Arm® Cortex®-A series application processor cores.

Another popular way to implement home security camera designs is an approach called "Light AI," which uses less powerful chips from manufacturers like Omnivision, Kneron and NXP Semiconductors. It uses sets. While Omnivision and NXP are already world-renowned suppliers of advanced sensor and processor chips, Kneron is rapidly emerging as a leader in AI technology, with its KL520 being EE Times' top 10 chipsets for edge AI applications. (www.eetimes.eu/top-10-processors-for-ai-acceleration-at-the-endpoint/).

So these AI-inspired home vision systems are effectively computer cameras. It is very different from traditional CCTV (closed circuit television) cameras or security cameras that simply record video footage with image timestamps (see Figure 1). As such, the new generation of home security cameras employs a computer-style architecture with high-speed DRAM memory large enough to store complex inference engines and other application code (Figure 2). reference)

Figure 2: A typical home security camera system architecture is equipped with ISP-compatible DRAM system memory (e.g. Ambarella CV25S)

However, unlike tablets, smartphones, and other general-purpose computers, home vision systems have specific image and inference capabilities with optimized components and system architecture. This means that DRAM is used specifically for the needs of image processing functions, rather than for general computing tasks.

As a result, home security camera manufacturers have found it beneficial to use low- to medium-capacity LPDRAM instead of the high-capacity 4Gbit, 8Gbit, and 16Gbit​ ​DRAM used in smartphones and laptops. I'm beginning to recognize

Figure 3: Kneron KL520 SoC demo board. KL520 NPU (artificial intelligence purpose-built processor) ranked among top 10 chipsets for edge AI applications by EETimes

Low power consumption and long-term product supply

This has a big impact on how manufacturers decide which DRAMs to buy from which suppliers.

For high-end computers and tablets, OEMs need to maximize DRAM memory capacity and bandwidth while minimizing cost, so they must use the latest DRAM technology manufactured with state-of-the-art processes. .

Low power consumption-For example, smart doorbell designs require long operating times for disposable alkaline batteries to minimize the frequency of battery replacement.
Longevity-The economics of the home vision system market require that a commercialized design be on the market for three years or more. For this reason, component suppliers of these products should commit to long-term supply so that OEMs do not have to redesign their hardware at EOL of their parts.
Reliability-Home vision systems are carefully considered and fixed to capture the desired scene. This means that repairs or replacements require a technician to visit the installation site, which is far more expensive than shipping a broken laptop to a repair shop. Customers and OEMs therefore focus on the reliability of individual components and the overall system.

This determines the key factors OEMs use to evaluate DRAM products and suppliers. On the other hand, we can see that the approach of specialty DRAM manufacturers contrasts sharply with that of the leading DRAM manufacturers focused on the PC and server markets, which are manufactured on cutting-edge processes.

Specialty DRAM suppliers offer a wide range of products for applications that can work with legacy DRAM technology rather than state-of-the-art products. In addition to standard versions SDRAM​ ​DDR, DDR2 and DDR3, Winbond Electronics offers low power DDR​ ​LPDDR2, LPDDR3 and LPDDR4/4x Mobile DRAM for power sensitive designs.

Mainstream DRAM manufacturers' manufacturing processes are optimized for memory capacity and die cost. Manufacturers who have not followed the cutting edge manufacturing process have the freedom to improve their traditional manufacturing processes to better meet customer needs for the industrial, automotive, medical and consumer markets. Winbond Electronics uses a low-power process for mobile DRAMs in its own DRAM wafer fab in Taiwan.This technology enables even standard DRAMs to deliver superior power performance over competitors' standard DRAM products. is possible.

OEMs operating in the home vision system market must also consider long-term supply of components. Again, specialty DRAM suppliers can meet market requirements. Winbond Electronics provides a standard long-term supply guarantee of 10 years from the date of market introduction of the product.

Finally, mainstream high-capacity DRAM has an interface compatible with the x86 microprocessor architecture used in PCs and servers, while specialty DRAM can be used with other types of processors, especially Arm architecture devices. can be easily connected with

To facilitate system integration for home vision system manufacturers, Winbond Electronics works with ISP chipset manufacturers to support reference design boards that provide a starting point for many security camera design projects. Software and hardware compatibility make system integration easier, especially for Ambarella ISP users.

Support rapid development of new, more advanced vision systems

With the evolution of AI and machine learning technology, there is a growing demand for higher signal processing throughput and faster memory to support ISPs. The specialty DRAM roadmap is introducing new standard technologies such as LPDDR4/4x to increase bandwidth.

Winbond Electronics' DRAM and Flash memory products, with the advantage of well-documented and third-party certified high-quality in-house manufacturing, enable OEMs to integrate intelligent new features into home monitoring and security vision systems. I will support you.

Click here for Winbond's DRAM products, which have strengths in long-term supply

Inquiry

If you have any questions regarding this article, please contact us below.

Back to Winbond Manufacturer Information Top

If you want to return to Winbond manufacturer information top page, please click below.