​A number of major Korean banks have begun implementing facial recognition as an auxiliary means of identification both in person and online.  Each bank seems to have chosen a path and timetable but the ultimate goal is to replace ‘private authentication’ or identification certificates that are used for identification.  Facial recognition applications have become more common in Korea where it is used for at commutation egress points and certain kiosk services, with biometric sensors, such as fingerprint or palm readers becoming a point of bacterial or viral transfer, while facial recognition requires no physical contact and in some cases is less expensive.
Hana Bank (086790.KS) has already registered 1.94m users in its facial database and is certifying ~10.6m customer transactions each month with the system, with the bank estimating that ~20% of their customers are using the facial recognition system to replace passwords, fingerprints, or ID certificates.  Daegu Bank  (139130.KS), that had previously used a combination of an ID card and a video call to verify remote customer transactions, now uses a facial recognition system that compares a real-time video image with the picture on an ID card to verify authenticity and expects to expand the system to in person transactions that have been certificate-based in the past, and Shinhan Bank (SHG) pushed facial recognition to its offline branches to verify transactions without an ID card.
The Korean Fair Trade Commission, the same organization that unraveled a digital rights price fixing scandal involving 9 companies, including SK Telecom (SKM), Sony Music (SNE), and Korea Telecom (KT), has been the organization that has issued the abovementioned 'certificates' of identity, but is now looking to add facial recognition to its distributed ID services that will combine the technology with block chain for additional protection, but have stopped short of using the technology by itself, as the fear of ‘deep fakes’ has not been fully eliminated from the system.  While there will always be questions about the ultimate security of the data produced by such systems, in the cases above, entry into the systems is completely voluntary, as opposed to traffic and pedestrian system used in other countries.

​While an official announcement has not been made by Samsung Electronics (005930.KS) concerning the location of the company’s next semi fab, it seems they have already made an internal decision to build the new line at their existing complex in Austin, Texas.  South Korean trade press has reported that the company will begin construction of the new fab in 3Q of this year with a goal of mass production in 2024.  The official announcement is expected to be made around the time of the US/South Korean Summit, which is to take place later this month (tentatively May 21).  Samsung has been buying additional land near its existing fab since 2018, has accumulated a bit over 300 acres to date, and requested rezoning for the land.
The new line is expected to be a 5nm EUV process line that will cost ~$17.6b US and will be in addition to the S2 fab that Samsung has in production in Austin, which has a capacity of 92,000 wpm and is primarily working at 14nm.  Samsung has already sent staff to Austin to supervise the construction of the new fab, according to Korean press, which leaves Arizona, and upstate New York, the two other potential locations for the fab, out of luck.  Arizona had been offering a $9,000 tax credit for every job created by the fab and New York was offering $900m in funding to develop the site if Samsung were to choose that location.  The original Austin plan requested by Samsung was for $718.m from Travis County and $87.2m from the city of Austin in tax breaks over 20 years, with construction to begin this quarter and mass production in 3Q 2023.  If the Korean press is correct, there would only be a small delay from the original plan, although the financials might have changed.

What is normally an excursion out to sunny California, the 2nd virtual SID (Society for Information Display) conference begins today.  Historically the show has been a 3 day+ event where technical papers and talks concerning various aspects of the display industry are presented, with an exhibition that showcased companies both large and small and gave visitors the opportunity to see new or potential products.  As of last year, the show became virtual and no longer gave the opportunity to wander through the exhibits and ask questions to staff and management so we take our virtual walk-thru a bit more selectively and pick a few topics to focus on as we ‘walk’ the virtual aisles.
While announcements from display and related companies will be made over the next two days, Samsung Display (pvt) was one of the first out of the box, showing off its S-Foldable prototype, a double folding OLED panel, a tablet/notebook size foldable that unfolds to become a 17” monitor, and the slide-bar foldable that we have mentioned in the past.  No dates about if and when these displays will make it into commercial products, and we note that Samsung Display is only the supplier of the displays and not the producer of actual CE products that might include or use them, so actual releases would be dependent on the brand, which in all likelihood would be Samsung Electronics, with an over 80-% share of the small/medium OLED display market.  None of these prototypes are new in that they have been seen in IP or other demos, but Samsung has promised to expand its line of foldable devices this year, which almost implies that at least one or two are on a path toward commercial production.

​In what seems to be a battle between a controversial Chinese equity manager and a number of listed Chinese companies, Chinese panel producer Visionox (002387.CH) has received a request for clarification of information by the Chinese Securities Regulatory Commission.  While the details of what these companies are being accused of is still a bit unclear, it seems that it has to do with whether the companies are ‘managing’ market value through JVs or other alignments with 3rd parties that have not been disclosed.  The request from the CSRC asks the company to respond and do a ‘self-examination’ as to whether there is any information that should be disclosed that has not been disclosed, which is a rather odd request of a company that has been accused of non-disclosure.
That said, the accusations about market manipulation are just that and no hard and fast information has been given to the public about such manipulation, but the posting caused the stock to drop by about 6% in what was a rather weak market for display stocks in China.  The requested ‘self-examination’ results are due for review by the CSRC today, and should be reviewed quickly, with the results, particularly a current list of shareholders to be made public after the review.

Earlier this month (05/11/21) we indicated that smartphone shipments in China dropped 23.8% m/m and 34.1% y/y after a strong 1Q, shining some light on the very aggressive shipment targets that Chinese smartphone brands had set for this year.  While last year was certainly an anomaly, this year seems to be at least heading toward a bit more consumer normality, at least in some regions, and while the unusual spates of pent-up demand last year might have given hope to Chinese smartphone brands that we were entering a new phase of smartphone demand, this year the Chinese smartphone space is facing component shortages and a currently severe outbreak of COVID-19 in India, the 2nd largest market for a number of Chinese smartphone brands.
At least four Chinese smartphone brands have reduced their previous 2021 shipment targets, citing shortages and issues in India, but regardless of the reason, the reductions are being passed down to component suppliers.  Of course, even with the reduced expectations for 2Q, Chinese smartphone brands are expecting an increase in 3Q and 4Q as new models are released, although no mention of what the strength will be based on or how the same issues that will affect shipments in 2Q will be dealt with in 2H. The question remains as to whether these reductions will give component suppliers a more realistic view of actual demand, rather than overbooking or component stockpiling, which has been an issue for the industry, particularly this year.  The table below shows the changes made to shipment targets from four Chinese brands based on current targets and those made in 4Q last year. 
Missing are Huawei (pvt) whose trade issues with the US have already severely impacted the company’s shipments both in China and globally, which would have been reduced regardless given the sale of its Honor (pvt) brand, along with Oppo (pvt) and Vivo (pvt) who have yet to update targets.  With component shortages and depleted inventory levels, Huawei’s shipment levels and share have become far less relevant, with others Chinese brands vying to fill the gap.  We note that the parent organization of Oppo, Vivo, RealMe (pvt), and OnePlus (pvt) is privately held BBK Electronics (pvt) based in Guangdong, China.  

​According to China’s Ministry of Industry and Information Technology, as of last week China has built more than 819,000 5G base stations, or about 70% of the total global count, and mobile phone user connections have reached 280m, more than 80% of the total global connections.  While patent declarations do not have intrinsic value, the organization also noted that China was the leader in 5G ‘standard essential patent declarations’, increasing share by almost 5% y/y.

China’s largest foundry SMIC (688981.CH) reported stronger than expected 1st quarter sales of $1.1b, up 12.5% q/q and up 22% y/y and gross profit of $250.1m, up 41.5% q/q and up 7.1% y/y, with gross margins increasing from last quarter’s 18.0% to 22.7% in 1Q, although down from the 25.8% GM seem in 1Q ’20.  The company shipped 1.559m 8” wafer equivalents, up 10.1% from the previous quarter and up 10.8% y/y at a utilization rate of 98.7%, up from 95.5% in 4Q and up slightly from 98.5% seen in 1Q last year. The company expects sales to increase by 17% - 19% in 2Q and gross margins to increase to between 25% and 27%.  SMIC had given full year guidance in February, expecting sales to increase in the mid to high single digits and for gross margins to be in the mid-teens, now indicating that they expect to beat that guidance for the year, but given the uncertainties in the industry, they declined to give a new forecast.  The company pointed to a number of product segments that were out of stock currently, including power management, RF, drivers for MCUs, CIS image sensors and certain storage products, and expected that 15nm, 18nm, and 40nm nodes would be the tightest.  SMIC also indicated that it has tried to migrate some products to less tight nodes, but in reality the overall capacity remains constant and that at such high utilization rates, the migration results are relatively small.
While the good news certainly was good, as this was the company’s 2nd best quarter ever, management did indicate that while negotiations were heading in the right direction, it was expecting that the US restrictions placed on the company last December by the previous administration, could delay the mass production timeline of the company’s new $2.35b 28nm/40nm foundry project in Shenzhen that was announced in March and could also delay the planned expansion of its 8” and 12” fabs, which were expected to see 45,000 wpm and 10,000 wpm increases this year.  With expected delays, the expansion contribution will be pushed to the back end of 2H and will likely be less than expected for the full year because of the delays in procuring equipment.  While that is certainly distressing for SMIC, they still gave 2021 guidance a boost, even under these circumstances.
Given that SMIC is the 4th largest global foundry, despite the political optics of lessening the hold the US has over equipment vendors that supply SIC, the global semiconductor shortage affects the US economy in many visible and some less visible ways.  Automotive plant shutdowns, higher device prices, and potential end-user product shortages will capture more headlines than allowing SMIC to build capacity.  While eventually equipment trade license grants would come to light and be promoted by anti-China trade groups, granting trade licenses, especially for 28nm equipment, would be a more subtle way of helping to alleviate the semiconductor shortages without making visible public policy changes.

While power outtages in the semiconductor space are rarely good news, the power outage seen across Taiwan yesterday seems to have been one that caused little damage.  The rolling blackout that occurred on Thursday evening affected between 4m and 6m homes out of almost 9m did reach some fabs operated by Taiwan Semi (TSM), United Micro (UMC), Powerchip (6770.TT) and Vanguard International (5347.TT), but officials have said the impact was negligible or minor, while AU Optronics (AUOTY) and Innolux (3481.TT) reported normal operations at their plants on the island.  Semiconductor test and assembly provider ASE Technology Group  (3711.TT) however reported that it was ‘slightly affected’, and ChipMos (IMOS) indicated that some of its IC driver lines were ‘more or less’ impacted.  Taiwan foundries account for ~56% of 12” wafer production, ~40% of 8”, ~21% of FRAM production and 1% of NAND.
The outage was caused by a sub-station fault at Taipower’s (state) Hsinta Power Plant, a coal and gas fired plant in southern Taiwan, which caused the grid to have insufficient power for the entire island at which point officials shut down the plant and instituted rotating blackouts, allowing full power for groups of ~ 2m citizens each to allow for people to get off elevators and other electrical conveyances.  Taiwan is also going through both a heat wave and a serious drought, which has reduced power availability from hydroelectric generation by ~16% y/y.

​As mini-LED backlights become more popular (and available) and more complex, the video processor in your TV has to get smarter and each TV brand has their own flavor of what the video processor should do and how it should do it.  On a general basis, the video processor in a TV takes an incoming signal and looks at various characteristics of the signal to create a better picture, but they have become far more important and complex as TVs have also become more advanced.  Originally the video processor only had to take the signal and break it into two streams, one for the even rows on the display and one for the odd rows.  This process, called interlacing, was used to allow the display (CRT) to paint the odd rows before the even row phosphors had a chance to dissipate the image rather than the system painting the entire screen at once, which would allow the image to fade before it was entirely redrawn (flicker).
With the advent of LCD displays, which require backlights, the purpose of the video processor began to change.  In the early days of LCD TVs, the fluorescent backlights were always on, leaving it up to the liquid crystal to ‘close’ when the image needed a black pixel.  Unfortunately liquid crystal does not close completely, which causes blacks to be ‘grayish’ as some of the light leaks through.  Further, in cases where a dark pixel is next to a light pixel, the brightness of the backlight from the light pixel leaks into the dark pixel and causes what is known as ‘bloom’ or halos around bright images.
When fluorescent backlights were phased out and replaced by LEDs, set designers realized that instead of placing LEDs around the edges of the display (edge-lit) and leaving them on all the time, they could place them behind the display (direct-lit) and eventually realized that they could control those LEDs and reduce the negative effects of bright LEDs by dimming or by turning them off (local dimming).  This ushered in a new level of complexities for video processors, as there had to be some way for the system to anticipate which LEDs should be dimmed at what points and to what levels.  Since video processors were already looking at the details of the video signal, designers took that information and translated it signals that would control the backlight LED brightness.  This was a great idea, but with only a relatively small number of LEDs in the direct-lit backlight, the same ‘gray’ and ‘bloom’ problems still occurred.
As the LED industry became more sophisticated and LEDs became smaller, designers began to increase the number of LEDs, which when group together are called zones.  Only a few years ago high-end TV sets had 10 or 20 zones, which increased the contrast and reduced bloom, but as OLED TVs became popular, where each pixel was individually controlled, even direct lit dimming systems could not compete.  In order to increase the granularity for LCD TVs, designers kept adding LEDs and zones, but along with that control came a need for more image processing granularity and more sophisticated video processors.
Video processors now can look at an image and break down that information into a map of every pixel.  As processing speeds have also improved, the processor can evaluate not only a single pixel, but what the individual pixels around are doing at the same time and can better evaluate how it should tell the LED backlight to work.  That said, there are still many more pixels than LED backlights in even the higher tier TVs, so there are still evaluations that the processor must make to figure out what looks best for each frame.  Mini-LED backlights will gin up this game even further with much higher LED and zone counts, but with each increase the video processors algorithms must become more sophisticated (‘smart’), with AI being the most recent buzzword as to how to add that increased video processing intelligence.
AI, at least in this case, is really a system that looks at each time it evaluates a frame and ‘remember’ what it saw.  It keeps compiling that data and based on that, gets an understanding (learns) what the pixels surrounding each other would be doing in a particular circumstance, so when it sees a familiar pixel pattern, it does not have to evaluate the entire image and sets the LEDs in the way it has learned is most common.  Of course this leaves the success of such systems in the hands of programmers, who have to figure out what the most important information is and how to evaluate it, but using Ai allows the video processor to control more LEDs with more granularity.  Whether that leads to a much better picture is still a bit subjective but in theory makes sense.
As the consumer becomes more familiar with what passes for AI in CE devices, designers are faced with the ‘What have you done for me lately?’ question from marketing and Sony (SNE) among the leaders in video processing, has taken things to the next level.  Sony’s most recent iteration of their Bravia™ TV line is touting the next level in video processing, calling it ‘cognitive intelligence’.  While just another buzzword to confuse consumers, the company says the new processor looks at the ‘whole image’ rather than on a pixel by pixel basis and breaks it down into ‘zones of concentration’ (our words) that it feels deserve attention.  This could mean that it tries to understand what the key images in a frame are and works toward optimizing that section, rather than spreading it processing power across the entire frame.  Sony says this mimics the way a human would see the image, focusing on what is most important.  As most media display evaluation are subjective, and most marketing literature is less than empirical, we would have to wait to see actual real-world results to judge, but at the least it seems to be a step in the right direction in terms of a more practical way of looking at the problem, and it will also serve to separate Sony’s marketing from the rest of the ‘AI’ crowd, likely as important as the improvement in technology. The 45 second promo video is below:
https://youtu.be/-EhB7dUJ29g
 

​Periodically we check to see what companies in display and related areas are actively procuring equipment, giving some indication as to the progress of either expansion plans or new projects.  In April the most aggressive purchaser was BOE, who was accepting bids for a variety of equipment for its OLED fabs in Ordos and Mianyang, and LCD equipment for its Gen 10.5 fab in Hefei, and more major purchases for the new G6 OLED fab BOE is building in Chongqing.  Tianma (000050.CH) was also a significant purchaser for their OLED fab in Xiamen, which is under construction, and HKC (190.HK) for its H5 LCD fab, which is expected to open later this year.
We note that there were a number of procurements from an unnamed micro-OLED display project, which we tracked down to be a project initiated back in 2017, involving BOE, a number of Chinese development funds, US based Kopin Corp (KOPN), and a company known as Olightek (pvt) (aka Yunnan Invensight Optoelectronics), that is controlled by China North Industries Group  (state), also known as The China Ordinance Industries Group or Norinco, (state), one of the largest global defense contractors.   
The project is developing an OLED micro-display production line, in addition to Olightek’s existing production line and is based on the same technology.  The line is developing a silicon based micro-OLED display that has purchased equipment from South Korea’s Sunic Systems (171090.KS) toward that end late last year.  BOE began offering a 0.39”OLED micro-OLED display in late 2019 and a 0.71” full HD micro-OLED display last year that has over 3,000 pixels/inch.   BOE is also involved in a JV with Rohini (pvt) that is developing a micro-LED backlight and direct emission system.