STMicroelectronics. 3D Sensing, an opportunity yet to be priced in INDEPENDENT RESEARCH UPDATE. TMT Fair Value EUR13.7 (price EUR12.

Transcription

1 INDEPENDENT RESEARCH UPDATE 1st February 2017 STMicroelectronics 3D Sensing, an opportunity yet to be priced in TMT Fair Value EUR13.7 (price EUR12.20) BUY Bloomberg STM FP Reuters STM.FR 12-month High / Low (EUR) 12.5 / 4.6 Market capitalisation (EURm) 11,110 Enterprise Value (BG estimates EURm) 10,881 Avg. 6m daily volume ('000 shares) 2,209 Free Float 70.3% 3y EPS CAGR 52.5% Gearing (12/16) -11% Dividend yield (12/17e) 2.40% YE December 12/16 12/17e 12/18e 12/19e Revenue (USDm) 6,972 7,748 8,321 8,744 EBITA USDm) ,065 Op.Margin (%) Diluted EPS (USD) EV/Sales 1.65x 1.52x 1.36x 1.24x EV/EBITDA 11.5x 9.1x 6.1x 6.0x EV/EBITA 37.4x 17.0x 11.6x 10.2x P/E 46.0x 20.9x 14.4x 13.0x ROCE /07/15 31/10/15 31/01/16 30/04/16 31/07/16 31/10/16 31/01/17 STMICROELECTRONICS (PAR) SXX EUROPE 600 At the STMicroelectronics earnings publication on Thursday 26th January, the group communicated information that added weight to our idea that it has won a positioning in the next generation iphone with its Time-of-Flight technology (ToF, 3D Sensing). The accumulation of factors converging around this idea has therefore prompted us to include the assumption in our model. Given the significant leverage to margins, pricing in this opportunity prompts us to raise our EPS estimates by 37% on average over three years. We are reiterating our FV of EUR13.7 and our Buy recommendation adopted on 30th January. Converging information adds weight to our assumption of a design-in for the next generation iphone. The first news item that suggested a design-in at Apple dates back to July 2016 when the Dauphiné unveiled that Apple had just opened an R&D centre in Grenoble, STMicroelectronics' stomping ground. During our research on Soitec, we then noted the interest that Apple could have in leveraging ST's proprietary technologies (ToF, mirco-mirrors, FD-SOI ) and its expertise in imaging, to develop a 3D sensor and offer the iphone the devices necessary to offer a convincing augmented reality experience. These assumptions were backed in November 2016 by magazine Challenges, which revealed that ST was going to work for Apple in image sensors (without providing further details). Finally, during the conference call on 26th January, ST provided several elements that added weight to our assumption. Among these was the acceleration in investment spending in fabs, justified by the prospect of a sizeable opportunity as of H A share price that does not fully reflect the incremental opportunity. Despite the share's healthy performance in 2016, we believe that it is still an investment opportunity worth seizing. The incremental opportunity prompted by ToF sensors does not seem to be fully reflected in consensus figures or the share price. As such, on the basis of 1/ buoyant momentum in the short term, 2/ further improvement in the group's fundamentals and 3/ average EPS growth over three years of 52.5% after integrating additional volumes in ToF sensors as of H2 2017, and which point to a 2017e PEG of 0.4x (2017e P/E of 19.4x), we are reiterating our Buy recommendation adopted on 30th January. Analyst: Sector Analyst Team: Dorian Terral Richard-Maxime Beaudoux 33(0) Thomas Coudry dterral@bryangarnier.com Gregory Ramirez r r

2 Company description STMicroelectronics is a Franco-Italian manufacturer of semiconductors. The group has a broad product portfolio that spans from power management components to integrated circuits for industrial sector, automotive and consumer applications. Declining for several years, the group currently executes a transformation plan to restore growth and positive margins. Simplified Profit & Loss Account (USDm) 31/12/14 31/12/15 31/12/16 31/12/17e 31/12/18e 31/12/19e Revenues 7,404 6,897 6,972 7,748 8,321 8,744 Change (%) -8.4% -6.8% 1.1% 11.1% 7.4% 5.1% EBITDA 1, ,003 1,295 1,843 1,817 Depreciation & amortisation Adjusted EBIT ,065 EBIT ,065 Change (%) -% -35.0% 96.3% 188% 58.9% 9.0% Financial results (19.2) (22.0) (21.0) (19.4) (20.8) (21.9) Pre-Tax profits ,052 Tax (33.0) (122) (165) (165) Profits from associates (43.0) Minority interests (0.60) (6.0) (6.0) Net profit Restated net profit Change (%) -% -24.7% 54.9% 119% 43.7% 10.9% Cash Flow Statement (USDm) Operating cash flows ,090 1,679 1,640 Change in working capital (76.0) (74.9) (132) (97.9) Capex, net (496) (467) (625) (1000) (749) (743) Financial investments, net (288) (49.0) Dividends (357) (356) (251) (280) (349) (349) Issuance of shares Issuance (repayment) of debt 774 (200) Other (156) 0.0 (22.0) Net debt (546) (494) (513) (248) (696) (1,145) Free Cash flow Balance Sheet (USDm) Tangible fixed assets 2,647 2,321 2,287 2,683 2,566 2,557 Intangibles assets & goodwill Investments Deferred tax assets Current assets 2,700 2,570 2,518 2,690 2,882 3,024 Cash & equivalents 2,351 2,106 1,964 1,699 2,147 2,596 Total assets 9,008 8,191 8,008 8,311 8,834 9,416 Shareholders' equity 5,055 4,693 4,596 4,801 5,265 5,804 Provisions Deferred tax liabilities L & ST Debt 1,805 1,612 1,451 1,451 1,451 1,451 Current liabilities 1,477 1,266 1,412 1,509 1,569 1,613 Total Liabilities 9,008 8,191 8,008 8,311 8,834 9,416 Capital employed 4,509 4,199 4,083 4,554 4,569 4,659 Ratios Operating margin Tax rate NM NM Net margin ROE (after tax) ROCE (after tax) Gearing (10.80) (10.53) (11.16) (5.16) (13.22) (19.73) Pay out ratio Number of shares, diluted Data per Share (USD) EPS Restated EPS % change -% -23.9% 53.9% 120% 45.4% 10.9% EPS bef. GDW NM NM NM NM NM NM BVPS Operating cash flows FCF Net dividend Source: Company Data; Bryan, Garnier & Co ests. 2

3 Table of content , the year of augmented reality A fundamental trend The situation heading into Indications from Apple thanks to the patents filed by the group Google is also very active Baidu already has projects in production and Time-of-Flight sensors Proximity sensors, existing technologies Time-of-Flight is clearly capable of measuring the distance between two points An application range not limited to augmented reality Significant impacts for ST Factors adding weight to our view that ST is set to benefit from this technology alongside Apple A sizeable sales opportunity A significant impact on the group's margins Breakdown of our model (as of 30th January) Valuation SOTP DCF Sensitivity of our FV to the ToF sensor ASP We reiterate our Buy recommendation Bryan Garnier stock rating system

4 Today, the rare examples of augmented reality (AR) that exist are the result of a somewhat unconvincing development, given the lack of financial means available to developers , the year of augmented reality A fundamental trend Today's smartphones are ever more performant, carrying out ever more daily tasks that make them vital objects in the eyes of all their users. However, for several quarters now, we have noticed a slowdown in this market segment due to a degree of saturation and also a lack of fundamentally new factors. And in terms of innovation, two new concepts indeed spring to mind: 1/ foldable screens that transform the smartphone into a tablet and 2/ a significant improvement in smartphone capacity to carry out augmented reality by using specific sensors. Today, the rare examples of augmented reality (AR) that exist are the result of a somewhat unconvincing development, given the lack of financial means available to developers. Indeed, smartphones are incapable of seeing the world in 3D and AR applications therefore have to make do with only gyroscopes, accelerometers and GPS systems to function. Despite everything, the interest in these technologies is very tangible. The best example remains the success of the Pokémon Go game, which attempts to insert virtual creatures into the real world. The mobile game application has nevertheless revealed two major problems: 1/ a somewhat unconvincing rendering in the majority of situations (see Fig 1.), and 2/ a high level of energy consumption due to the need to have complex algorithms working permanently. Fig. 1: The best example of AR today is Pokémon Go, a success despite a still hazardous and unconvincing rendering Expected rendering Actual rendering of Pokémon Go A sticker in the middle of the camera view, not aware off and unable to interact with elements of the real world. Source : Niantic To resolve these problems, we believe the use of Time of Flight (ToF) sensors could be the solution. To resolve these problems, we believe the use of Time of Flight (ToF) sensors could be the solution. These sensors are the only ones capable of measuring distances and hence enabling smartphones to perceive the environment in 3D, a prerequisite for obtaining convincing augmented reality. We can then go on to imagine a large number of applications ranging from ordering tailor-made suits via a mobile application to the integration of armchairs in a sitting room via a furniture sales application, and obviously including games on mobile handsets. 4

5 Whereas augmented reality (AR) was clearly wiped out by virtual reality (VR) barely a year ago, more and more applications prove that the market opportunity of AR is far higher The situation heading into 2017 Whereas augmented reality (AR) was clearly wiped out by virtual reality (VR) barely a year ago, more and more applications prove that the market opportunity of AR is far higher. Note that during the 2016 Mobile World Congress, numerous appliances dedicated to virtual reality were presented. Among others were the Gear VRs by Samsung, the commercial version of Vive by HTC and the 360 VR by LG. And to welcome these, Mark Zuckerberg made an astounding entrance to the congress, taking to the stage while all members of the audience were wearing a pair of VR glasses. Fig. 2: VR honoured at the 2016 MWC Source: Mobile World Congress During the second half of 2016, the focus gradually moved towards augmented reality. Whereas VR plunges the user into a virtual world, augmented reality aims more at improving our perception of the real world. So far, the most famous application of augmented reality is the mobile game Pokémon Go, which uses various sensors (gyroscope, accelerometer, camera, and GPS ) to make Pokémon creatures appear virtually in everyday places. Although very popular, the Pokémon application is fairly limited relative to the opportunities offered by the AR technology. In addition, a number of major groups have clearly shown their interest in this technology, like Apple, which in contrast has remained very discreet about virtual reality. 5

6 In November 2013, Apple unveiled its interest in augmented reality by acquiring an Israeli fabless group specialised in 3D processing, PrimeSense. Between 2014 and 2016, the publication of the PrimeSense/Apple patent showed that the Cupertino group was continuing to work on device control technologies using movement detection Indications from Apple thanks to the patents filed by the group In November 2013, Apple unveiled its interest in augmented reality by acquiring an Israeli fabless group specialised in 3D processing, PrimeSense. The name of the company probably means little to the majority of people, but the start-up company was at the root of the technology used in a well-known consumer device, namely the X Box Kinect by Microsoft. In its first version, which embedded the Prime Sense technology, the Kinect used a complex triangulation system implying a traditional camera, an infrared camera, an infrared projector as well as a complex algorithm. Apple spent around USD360m on this acquisition, which has so far not given rise to an embedded technology in an Apple device, whereas many observers were predicting that the PrimeSense technology would be embedded in the iphone 6. Between 2014 and 2016, the publication of the PrimeSense/Apple patent showed that the Cupertino group was continuing to work on device control technologies using movement detection. In addition, as we have already mentioned, the PrimeSense technology is based on analysis of a 3D scan of the object's environment. However, this technology and the associated material is perfectly well suited to augmented reality. In August 2016, Apple CEO Tim Cook started to discuss the core dimension of augmented reality in the group's vision. Mr Cook estimates that AR could be an "enormous phenomenon", thereby suggesting that it could represent a significant evolution in Apple products, as were multipoint capacitive touchscreens and inertial navigation units (gyroscopes and accelerometers) for the emergence of the smartphone. During the same event, Apple's CEO stated that the group was investing significant amounts in this technology and indicated an investment in R&D over more than five years. For Apple's chairman, the AR technology is far more easily accepted by consumers than virtual reality, which still requires 1/ financial means, bearing in mind that the VR headsets required are sold for around USD500, 2/ acceptance by consumers, which is not yet guaranteed given the headsets' lack of discretion and the need for powerful calculation tools (PC or console) to do more once users want more than a rapidly boring roller-coaster demo, and 3/ consumer applications, since while video games are the first candidates for adoption by VR, they remain an activity destined for a specific type of user and other applications fall rapidly into the professional domain, thereby limiting volumes by as much and hence the interest for developers to invest in this technology. Things are different for AR since the applications are easier to develop, as shown for example by cartography systems or casual games, since the implementation of AR in games is widely simplified relative to VR and requires fewer resources. 6

7 In November 2016, the specialised news website, Appleinsider, also published an article highlighting the publication of a US patent filed by Apple concerning the use of AR for its cartography application and Plan guidance. In November 2016, the specialised news website, Appleinsider, also published an article highlighting the publication of a US patent filed by Apple concerning the use of AR for its cartography application and Plan guidance. This detailed the schemes showing how an evolved guidance system works, with various information inserted into a view of the real world filmed in real time by the smartphone's camera. Fig. 3: Virtual reality schematised and patented by Apple Source: Appleinsider, via the US Patent and Trademark Office " AR I think is going to become really big. [ It] gonna take a little while, because there s some really hard technology challenges there. But it will happen. It will happen in a big way. And we will wonder, when it does [happen], how we lived without it. Kind of how we wonder how we lived without our [smartphones] today." - Tim Cook, Apple CEO Whereas expectations are now for Apple to deliver on subjects that are not gadget-based, AR could be a solution that would help the group to stand out again. In all, AR seems to be a natural technology for Apple, as touchscreens have become for smartphones. Whereas expectations are now for Apple to deliver on subjects that are not gadget-based, AR could be a solution that would help the group to stand out again. It is general knowledge that Apple is particularly patient when it comes to integrating a new technology into its products. Apple would above all like use of these new technologies to be natural (capacitive touchscreen, Siri, inertial scrolling etc). For this reason, we are not surprised to see the implementation of AR in Apple's products. 7

8 Beyond Apple, which always develops its projects in a very secretive manner, Google is pursuing its AR project, named Tango, started in Google is also very active Beyond Apple, which always develops its projects in a very secretive manner, Google is pursuing its AR project, named Tango, started in The project aims to roll out a platform aimed at grouping together all of the artificial vision applications (computer assisted vision). In concrete terms for Google, this is an Android integrated software brick that is capable of making the most of a range of sensors including movement sensors and above all Time-of-Flight sensors, that enable the generation of a 3D cartography of the smartphone environment. Via its dedicated website, Google gives two examples of how its technology can be used. The first illustrates the possibility of precisely measuring and integrating virtual elements into a real environment (a bed selected from the catalogue of a furniture reseller and then projected into a room in order to see how it would look). The second example focuses on a more fun usage with the creation of games interacting with the real world (virtual dominos laid on a real table, interacting with other real-world objects). Fig. 4: Measuring surrounding space to integrate contents in real time Source : Google In early 2017, this initiative remains limited since only two devices available to consumers integrate the material necessary to ensure compatibility with Tango. In early 2017, this initiative remains limited since only two devices available to consumers integrate the material necessary to ensure compatibility with Tango. These are the Lenovo Phab 2 Pro launched in August 2016 and the Asus ZenFone AR presented at the 2017 CES. For the momentum, no teardown (an in-depth study by components suppliers) has been made public, such that doubts remain over the supplier of depth sensors for these two devices. 8

9 The Chinese internet research giant started to work on AR more than two years ago Baidu has indicated it is already in close collaboration with several major groups such as Yum! (Taco Bell, KFC, Pizza Hut ), Lancôme, L Oreal and BMW Baidu already has projects in production The Chinese internet research giant started to work on AR more than two years ago, but only very recently built a centre dedicated to the development of services for education, health and tourism, or even marketing. There is an appetite for this technology; we are seeing rapid adoption by our partners in a range of industries. - Andrew Ng, Chief Scientist of Baidu and Head of Baidu Research In this respect, Baidu has indicated it is already in close collaboration with several major groups such as Yum! (Taco Bell, KFC, Pizza Hut ), Lancôme, L Oreal and BMW. In the case of L Oréal, an application stemming from the collaboration enables users to play with AR and the bottles produced by L'Oréal in order to obtain promotional offers for example. For Yum!, a concrete example of this joint work was presented at the end of December 2016, namely a smart order column destined for Chinese KFC restaurants that can analyse the consumer's face in order to suggest products, or even a full menu. The smart unit is capable of analysing various factors such as age, type and facial expression, and then integrate suggestions into an AR video flow sent to the user. To facilitate acceptance of this technology by customers, the smart units include augmented reality games including the generation of photos modified by stickers. At present, the games units have been installed in 300 restaurants in China, but only one test restaurant has the version of the unit where facial recognition is activated. Fig. 5: Baidu and KFC are developing a new "intimate" virtual host concept Source: TechCrunch Still using a Baidu application, Beijing metro users can see a slightly modified version of the city in order to show the city's historical doors, only the relics of which remain today. 9

10 While augmented reality applications are likely to take off soon, pushed by the sector giants as already discussed, technical limitations to their success still exist, especially the lack of dedicated and adapted components. 2. and Time-of-Flight sensors While augmented reality applications are likely to take off soon, pushed by the sector giants as already discussed, technical limitations to their success still exist. For augmented reality to be convincing, it is important that smartphones can analyse (sense) the surrounding 3D environment. The technologies available at present are restrictive since they are based on complex algorithms (including high energy consumption) and very limited proximity sensors. However, the Time-of-Flight technology opens numerous possibilities and seems to provide exactly the right answer Proximity sensors, existing technologies Functioning of current proximity sensors Technologies that enable the generation of a 3D view already exist, with the Microsoft Kinect the best example of this. At present, these technologies are based on an infrared (IR) sensor/emitter, which combined with an algorithm, helps rebuild a 3D scene. Similarly, numerous mobile telephones have long embedded proximity sensors that de-activate the screen when the device is close to the users face in order to avoid accidentally hanging up during the call. Current proximity sensors are made up of an infrared (IR) emitter, and a receptor which is responsible for measuring the emitter's light return in real time. While these partly respond to a need, the sensors are nevertheless very limited. They function on the principle of light intensity of an infrared wavelength. As such, a current proximity sensor is made up of an infrared emitter, and a receptor which is responsible for measuring the emitter's light return in real time. If an object is sufficiently close to the emitter, then it automatically reflects the light ray, which is returned to the sensor. As such, if the sensor detects a sufficiently powerful radiation, this means that an object is fairly close. As of a certain threshold of light intensity, the receptor will therefore trigger an action (turning off the phone screen for example). The technology embedded in the Kinect uses a more complex triangulation system, based on several sensors and which requires substantial calculation power. Fig. 6: Current proximity sensors use the amplitude of a signal Source: STMicroelectronics 10

11 Paradoxically, with the most simple proximity sensors such as those generally embedded in smartphones, measuring distance is impossible Limitations of IR sensors Paradoxically, with the most simple proximity sensors such as those generally embedded in smartphones, measuring distance is impossible. Since these sensors use the amplitude (power) of a light signal, the reflection of the object located opposite the sensor will have a significant impact on the result. For example, a mirror placed a few metres away will trigger a sensor more rapidly than a mat black surface just a few centimetres away. This is why, according to ST, that some people with very dark hair often have problems of accidental hanging-up during the call. Indeed, the reverberation from black hair is so low that once the smartphone is positioned close to the ear, the sensor located near the hair detects an overly low light signal and therefore considers there is no obstacle nearby. In addition, current distance sensors are also very sensitive to ambient light Time-of-Flight is clearly capable of measuring the distance between two points Like traditional proximity sensors, ToF sensors are made up of two elements, the first being a light source (laser), the second a receptor (sensor) Functioning of ToF sensors Like traditional proximity sensors, ToF sensors are made up of two elements, the first being a light source (laser), the second a receptor (sensor). The resemblance stops there since the ToF sensor uses the time it takes for the photons to travel to calculate the distance with the first object the laser encounters. Fig. 7: ToF sensors measure the time taken by a photon to travel a distance between the emitter and the sensor Source: STMicroelectronics Compared to IR proximity sensors, ToF sensors are not sensitive to the reflectance of the target and are able to provide a precise measure of the distance The first advantage of this method is that it is not sensitive to the reflectance of the target. The second advantage is that it helps provide a precise measure of the distance. Indeed, the simple division by two of the time taken by the photon to travel between the laser and the receptor, multiplied by the speed of light, helps deduce the distance. 11

12 Fig. 8: An efficient means of measuring distance Sources: CERN, STMicroelectronics The receptor of a ToF sensor is made up of one or more photodiodes called Single Photon Avalanche Diodes or SPADs. The receptor of a ToF sensor is made up of one or more photodiodes called Single Photon Avalanche Diodes or SPADs. These can be produced on Complementary Metal Oxyde Semiconductors or CMOS wafers (the traditional manufacturing process for digital chips) and therefore be juxtaposed with the digital circuit which is responsible for counting the photons and measuring the arrival time, followed by the digital output, i.e. comprehensible for a processor. Fig. 9: SPAD architecture Source : CERN 12

13 In order to improve the performance of ToF sensors and especially their resolution, it is possible to multiply the number of diodes in order to create a SPAD grid. Fig. 10: The multiplication of SPADs helps improve the performance of the sensors 32x32 SPAD array Source : CERN Limitations of current ToF sensors While the use of SPADs is practical in terms of integration since it implies a CMOS construction (and hence a smaller sized chip and lower production costs), it still presents certain limits. The first is the very construction of the diode: once the diode is triggered by a photon, it requires time to return to its original state. This is dead time during which the diode is no longer active, and if other photons arrive they cannot be counted (high loss rate). In addition, the multiplication of SPADs requires a degree of uniformity in the dead time in order to ensure a correct performance of the component. Fig. 11: SPAD, dead time to control SPADs suffer of a dead time to be controlled in order to enhance the performance of sensors. Source : CERN, STMicroelectronics SPADs cannot store photocharges proportional to the number of photons received as charge-couple devices more complex and costly to produces - (CDDs) can. Since SPADs are dynamic units, a digital impulse is automatically generated when a photon is detected. In fact, SPADs cannot store photocharges proportional to the number of photons received as charge-couple devices more complex and costly to produces - (CDDs) can. ToF and hence the distance must therefore be calculated immediately via an associated digital circuit. This is a manageable problem as long as the number of pixels is not too high, but becomes problematic as the number of SPADs increases (in order to improve the definition of the sensor). Indeed, one of the methods of traditional reading used to read CMOS image sensors, Random Access Readout, rapidly becomes inefficient for ToF sensors since the system has the main default that it can only read 13

14 information from one pixel at a time (before moving to the next one) and the information from the other pixels is therefore lost. Since the functioning of ToF sensors is based on the measure of the speed of light and a perfect synchronisation with the emitter (laser), the results generated by this processing method rapidly become incoherent. Fig. 12: The Random Access Readout method used for CMOS image sensors is not optimal for ToF sensors Random Access Readout method Need for parallel Time to Digital Converter architecture Source : CERN While CMOS RAR sensors are fairly simple and hence cheaper to produce, they have the disadvantage of deforming images or measurements during motion, or in the case of a ToF sensor, during synchronisation with the laser. Fig. 13: Rolling Shutter vs. Global Shutter Example with a photo Global Shutter (CCD) Rolling Shutter (CMOS) Source: QImaging In these conditions, several other methods have been developed but none currently enables the creation of a high resolution, high frequency Time of Flight sensor. It is in this backdrop that we believe STMicroelectronics has a portfolio of technologies capable of resolving some of the sticking points and thereby offering a sensor 1/ of an acceptable size to be integrated into a smartphone, 2/ at an acceptable price, using well-mastered production procedures, and 3/ that is energy efficient in order not to penalise the autonomy of smartphones. 14

15 Current ST s ToF devices allow only measurement of the distance between two points STMicroelectronics: ToF + globalshutter + micro-mirrors, a possible winning combination to exceed these limits STMicroelectronics has already developed a ToF sensor available since H However, in the current state of play, this sensor does not enable reconstruction of a 3D scene, but simply the measurement of the distance between two points. A version of this sensor is actually currently in use in the iphone 7 as a simple proximity sensor, positioned on the front face close to the front camera (it manages the screen shutdown when the phone is held by the ear. Fig. 14: STMicroelectronics VL53L0, simple ToF sensor already used in the iphone 7 Source : Chipworks The group is not the only one mastering ToF technology has an edge over its rivals thanks to differentiating technologies According to the group, this simple version of the ToF VL53L0 sensor is currently used by 70 smartphone models. In addition, it is important to note that STMicroelectronics is not the only player to have developed a ToF sensor (max 12 measurement points) but we believe the group has an edge over its rivals thanks to differentiating technologies. Among the competitors that have a ToF technology, we have identified the small German company PMD Tec, and Heptagon, recently acquired by ams (note available here). To get over the limits described previously, we believe STMicroelectronics is working on a number of innovations both for the receptor and the laser. Firstly, concerning the receptor (SPAD), we believe ST is looking at a material improvement in performances. Under this framework, the fact that the group masters the SOI technology could well be a factor that sets the group apart. The use of SOI wafers notably helps produce charge-coupled devices (CCD) or very high-performance CMOS simply on the same die (piece of silicon), capable of sensing instantaneously all of the pixels in the sensor or rather from the ToF sensor, i.e GlobalShutter (vs. RollingShutter for traditional CMOS sensors, see Fig. 13). Studies undertaken in laboratories (IEEE) have shown the interest and potential of using FD-SOI wafers for developing CCD image sensors juxtaposed with a CMOS circuit. For ToF sensors, this means a perfect synchronisation between the emitter and the sensor and hence improved precision and resolution. Finally, it is also coherent to think that STMicroelectronics is looking to leverage its micro-mirrors technology. This is a miniature mirror (Micro-Electro-Mechanical Systems - MEMS) capable of reflecting several thousand movements a second. They can be used to direct the laser's light rays in order to generate a grid pattern necessary for the generation of a 3D view of the scene filmed. 15

16 2.3. An application range not limited to augmented reality Beyond augmented reality, other applications could rapidly emerge for ToF sensors, some being more direct such as their use in improving the smartphone cameras, while others usages such as in the medical field are likely to require more time before becoming tangible market opportunities An improvement in smartphone cameras with bokeh However, we believe that multiple applications exist for this sensor. The one that makes the most sense in the short term is use of the ToF technology to add a notion of depth of field in images generated by a smartphone photo sensor. Indeed, these photo sensors often suffer from an association with very compact optics, limiting the reduction of depth of field in order to highlight a subject. Fig. 15: Notion of depth of field and bokeh Wide (e.g. a smartphone shot) Narrow (e.g. a DSLR shot) Source: QImaging Thanks to improved ToF sensors juxtaposed with image sensors, smartphones should be able to capture the various levels of depth in a scene. It would then be possible to precisely apply an out-offocus (bokeh) effect to the background area while keeping the subject perfectly clear. 16

17 Gesture recognition We also note potential in movement recognition which would help control a smartphone only by movements for example, with no contact with the screen. However, to make sense, the sensor needs to be positioned at the front of the phone to be used for this type of application. Fig. 16: ST provides examples of possible applications in gesture recognition Source : STMicroelectronics Numerous other applications in auto and medical Beyond smartphones, we see a huge field of possibilities for ToF sensor usage. Indeed, these sensors could quite easily be used in cars in order to control the driver's state of fatigue or the attention they pay to the road. Thereafter, in a second stage, they could be used to improve the performance of Lidars (distance sensors) over short distances during parking for example. Finally, other applications exist in the biomedical field or even in man-machine interface since these sensors could help efficiently control a computer by using gestures. 17

18 3. Significant impacts for ST ST provided indications that added weight to our view during Q4 earnings call 3.1. Factors adding weight to our view that ST is set to benefit from this technology alongside Apple On 26th January, STMicroelectronics reported it Q4 results and used the opportunity to provide a few indications that added weight to our view that the group is set to benefit significantly from these innovations. To try and understand the reality hidden behind the few items communicated to the market by ST, we need to look back at the various announcements made since July The first item we noted was the setting up of an R&D centre and a white room by Apple in Grenoble, just next to STMicroelectronics (our note is available here). Rumours at the time stated that this R&D centre would be dedicated to imaging without providing more details. We pointed out that Apple and STMicroelectronics had already teamed up in the past, in particular for inertial systems (gyroscopes + accelerometers) for iphone, although ST was then ousted firstly by Invensense and then by Bosch. The groups are still in partnership, in particular for the inertial unit in the Apple Watch. Elsewhere, we also underscored the fact that a collaboration in imaging with ST would make sense since the Franco-Italian group boasts strong expertise in this field. Then, at the end of November 2016, we discussed another rumour stemming from the French magazine Challenges, stating that ST would handle production of the future image sensors (photo sensors) for Apple (note available here). The magazine even stated that production would be handled at Crolles 300mm. The question remaining hanging and to which the rumour did not reply concerned the interest Apple had in joining forces with STMicroelectronics for its image sensors. Before Q4 results, we already highlighted the possible co-development of such a product. During the conference call, the group said that capex acceleration is due to a strong opportunity to materialise in H2 (usual major period for Apple suppliers) and that investments will target 300mm manufacturing sites, i.e. the sites used for ToF devices production. In the meantime, our research published two days later (available here) on a specific Soitec product, the imager-soi wafer, led us to state that STMicroelectronics and Apple could work together on improved image sensors that would be able to leverage on ST s proprietary technologies in order to combine the Time-of-Flight technology with a classic image sensor, in order to obtain a detailed 3D representation of filmed or photographed scenes. Finally, these uncertainties took shape during the STMicroelectronics publication on 26th January. Indeed, management stated that it intended to step up investments in production tools, with capex guidance set at USD1bn to USD1.1bn for FY 2017e. This is virtually double the group's normal level of capex. The group then simply stated that these investments were destined to strengthen front-end production tools, especially for its Crolles 300mm plant, and also back-end activities. As such, the aim is to strengthen production for digital products since these are produced in 300mm and probably products that require specific attention during packaging since otherwise the group would not have stated that some of this rise in capex was destined for the back-end. And ToF sensors are indeed produced in 300mm and have a complex packaging. In addition, still according to the group, this acceleration in investments should help meet "an opportunity" that could materialise as of H2. This corresponds exactly to the major period for all Apple suppliers since the new iphones are presented in September and their production generally starts slightly before their presentation conference. As an example, Dialog (80% of sales with Apple) reported growth of 46% in H relative to H1. 18

19 3.2. A sizeable sales opportunity While this adds weight to the idea that STMicroelectronics is currently working with Apple on the development of a new generation of Time-of-Flight sensors, the role that ST could play still needs defining. And the role that ST has in this innovation is a key factor for the sensor's price (ASP) and hence the size of the opportunity. At present, the public price (available at a retailer) of a ToF VL6180X sensor, presented in 2014, is USD2.65 per unit for orders of more than 5,000 units (minimum volume enabling the lowest price possible). As explained previously, we believe the sensor developed for Apple is a far more sophisticated product and even if the volumes potentially ordered by Apple could be very high, it seems reasonable to estimate that the ASP would not be lower than USD3.5. On the other hand, Apple pays around USD12-15 for the photo sensor modules behind the iphone 7, including the image sensor but also the optical stabilisation and autofocus. While ToF sensors, in a more sophisticated version, could be similar to a photo sensor, we do not expect their prices to reach this level. In our view, it is reasonable to estimate that the ASP would be in the range of USD3.5 to USD10. Pending additional details that would enable us to fine-tune our estimates we have assumed a low-end ASP, i.e. USD3.5. As such, we are assuming that the components that ST supplies Apple will be billed at between USD3.5 and USD10. This is a significant price gap and for this reason, pending additional details that would enable us to fine-tune our estimates and which could be made available shortly, we have assumed a low-end ASP, i.e. USD3.5. Based on virtually stable iphone volumes in 2017 relative to 2016, i.e. close to 215m units, the incremental opportunity of a prospective contract for STMicroelectronics works out to USD750m on an annual basis. More precisely, we recently adjusted our model (note published on 30th January available here) to include 60m units sold in Q3 and 50m units in Q4. In addition, we also take account of 70 smartphone models that currently embed this technology which represents an average contribution of close to USD50m per quarter. 19

20 3.3. A significant impact on the group's margins The additional sales prompted by ToF sensors in our model has an automatic impact on the group's margin. We understand that that the sensors are to be produced partly in Crolles 300mm, namely the plant which is currently the least loaded (around 75% load) and which has a significant impact on STMicroelectronics gross margin. A significant impact on the group's margins thanks to the significant cut in unused costs on the group level and the positive leverage of a significant volume contribution. In 2016, these unused costs had a negative impact of 50 basis points on the group's gross margin. We have therefore changed our gross margin estimates for 2017 and the following years in order to take account of: 1/ the significant cut in unused costs on the group level, 2/ the positive leverage of a significant volume contribution. Indeed, STMicroelectronics is an IDM, namely it operates its production plants itself unlike fabless players (Dialog, Melexis, u-blox ), which outsource production to one or more foundries (TSMC, GlobalFoundries, Samsung, UMC, SMIC ). We have therefore lifted our gross margin estimate for 2017e from 36.7% to 37.3%. Since the R&D for these ToF sensors has already been widely financed, we have no reason to believe that the group will have to step up its R&D investments during In addition, management pointed this out during the conference on 26th January. As such, we have maintained an average level of operating expenses of USD550m for 2017e, pointing to an adjusted EBIT margin of 8.9% vs. 8% previously. For the following years, we expect operating expenses to rise at a similar level to sales growth. Fig. 17: Operating margin to benefit from these additional volumes Ajdusted EBIT margin (in % of sales) 14.0% 12.0% 10.0% 8.0% 6.0% 4.0% 2.0% Adj. EBIT - Prev. Estimates Adj. EBIT - New estimates Adj. EBIT margin - Prev. estimates Adj. EBIT margin - New estimates Adjusted EBIT (in USDm) 0.0% e 2018e 2019e 0 Source: Bryan, Garnier & Co. ests. 20

21 3.4. Breakdown of our model (as of 30th January) Fig. 18: P&L Average EPS growth of 52.5% [in USDm] Q17e 2Q17e 3Q17e 4Q17e 2017e 2018e 2019e CAGR 16/19e Sales % Seq. growth 1.1% -2.2% 3.2% 9.9% -3.4% 11.1% 7.4% 5.1% Gross profit Gross margin 35.2% 37.1% 36.4% 37.7% 37.9% 37.3% 39.9% 40.4% SG&A R&D Other exceptional gains Adjusted EBIT EBIT margin 4.4% 7.1% 7.1% 11.1% 10.0% 8.9% 11.8% 12.2% D&A Adjusted EBITDA Cost of net debt Profit from associates Gain from investments Adj. pre-tax profit Adjusted tax Tax rate -11.3% -19.0% -18.9% -17.2% -17.5% -17.9% -17.1% -15.7% Non-control. interest Adj. net profit % of revenue 3.6% 5.6% 5.5% 9.0% 8.3% 7.2% 9.6% 10.2% Adj. EPS (in USD) % Seq. growth 53.9% -25.3% +2.3% +77.7% -10.1% 120% 45% 11% Sources: Bryan, Garnier & Co. ests. 21

22 Fig. 19: A visible acceleration in investments in 2017e [in USDm] Q17e 2Q17e 3Q17e 4Q17e 2017e 2018e 2019e EBITDA Change in WCR Other Cash flow from operating activities Capex Free cash flow Acquisitions Other Cash flow used for investing activities Proceeds of LT & ST debt Repayment of LT & ST debt Dividend payment Other Cash flow from financing activities Total cash flow CTA (cumulative translation adj.) Net increase in cash Cash at beginning of period Cash at end of period Sources: Bryan, Garnier & Co. ests. Fig. 20: Debt [in USDm] Q17e 2Q17e 3Q17e 4Q17e 2017e 2018e 2019e Cash and cash equivalents Inventories, net Trade accounts receivable, net Other Total current assets Property, plant and equipment, net Long-term deferred tax assets Other Total non-current assets Total assets Trade accounts payable ST debt Accrued income tax Other Total current liabilities Long-term debt Reserve for pension Other Total non-current liabilities Total equity Total liabilities and Equity Sources: Bryan, Garnier & Co. ests. 22

23 Our valuation of STMicroelectronics is the equi-weighted average of two methods, DCF and SOTP. 4. Valuation Until now, we valued STMicroelectronics using an equi-weighted average of three methods: DCF, SOTP and historical multiples. In view of the significant fundamental changes recently made by the group, a valuation using historical multiples now seems less relevant. As such, our valuation of STMicroelectronics is now the equi-weighted average of the first two methods, DCF and SOTP. Fig. 21: Valuation of EUR13.7 per share Valuation of EUR13.7 per share Method Weight FV (in EUR) Upside DCF 50% % SOTP 50% % Average valuation % Source: Bryan, Garnier & Co ests SOTP Fig. 22: Peers taken into account by division Division ADG AMG MDG Others Pairs Infineon, NXP, Texas Instruments, ON Semiconductor, Renesas Analog Devices, Broadcom, Infineon, NXP, Texas Instruments Analog Devices, Broadcom, Infineon, Microchip, NXP, ON Semi, Renesas, Texas Instruments ams, ON Semiconductor Source: Bryan, Garnier & Co ests. Fig. 23: SOTP valuation Sales 2017e EBIT 2017e Multiples VE/Sales Multiples VE/EBIT Valuation VE/Sales Valuation VE/EBIT Average Valuation $/Share (A) Division ADG x 13.20x AMG x 13.90x MDG x 13.45x Others x 11.70x Total (B) Net Debt + Restatement Net cash (A)+(B) Discount (due to management transition in May 2017 and ongoing restructuring of STB business) 25% Fair Value (USD) 16.4 Fair Value (EUR) 15.6 Source: Bryan, Garnier & Co ests. This method values ST at EUR15.6 per share, an upside potential of 25%. 23

24 4.2. DCF Fig. 24: WACC of 11.1% WACC European risk-free interest rate 1.6% Equity risk premium 7.0% Beta 1.4 Return expected on equity 11.1% Interest rate on debt 2.5% Market capitalisation (USDm) 11,664 Net debt on 31/12/16 (USDm) -513 Enterprise value (USDm) 11,151 WACC 11.1% Source: Bryan, Garnier & Co ests. Fig. 25: DCF valuation and sensitivity table iusdm (FYE 31/12) 2017e 2018e 2019e 2020e 2021e 2022e 2023e 2024e 2025e 2026e Revenues ,321 8,744 9,653 10,555 11,430 12,256 13,012 13,678 14,233 Change (%) 11% 7% 5% 10% 9% 8% 7% 6% 5% 4% Adjusted EBIT ,440 1,498 1,540 1,564 1,570 1,557 Operating margin 9% 12% 12% 14% 14% 13% 13% 12% 11% 11% Tax Tax rate 17.9% 17.1% 15.7% 15.7% 15.7% 15.7% 15.7% 15.7% 15.7% 15.7% Net operating income after tax ,263 1,299 1,319 1,324 1,313 Capex, net As a % of sales 12.9% 9.0% 8.5% 8.5% 8.5% 8.5% 8.5% 8.5% 8.5% 8.5% Depreciation & amortisation As a % of sales 7.8% 10.4% 8.6% 8.5% 8.5% 8.5% 8.5% 8.5% 8.5% 8.5% WCR As a % of sales 14.5% 15.1% 15.5% 15.0% 15.0% 15.0% 15.0% 15.0% 15.0% 15.0% Change in working capital Free cash flows ,132 1,175 1,206 1,225 1,230 Discounted free cash flows Total discounted FCF e-2026e 5,306 Discounted Terminal value e+ 5,114 Enterprise value 10,419 WACC - Net debt on 31/12/ [in EUR] 10.1% 10.6% 11.1% 11.6% 12.1% Equity value 10,932 8% Nbr of diluted shares (m) % Valuation per share (USD) % Valuation per share (EUR) % Upside vs. current share price -3% 12% Source: Bryan, Garnier & Co ests. Op. margin This method values ST at EUR11.8 per share, or a downside potential of 3%. 24