Perspective, page-35

How are we sure that the 2nd vendor is Sony and not TowerJazz?  Where and when was this information announced?

Surely it makes more sense of previous announcements and of the prudent logistics arrangements that led to Tower being selected in the first place that the "second MEMS vendor" is more likely to be one of the other MEMS fabs owned by Tower Semiconductor.  

To quote from the announcement of 19/7/16:

“... TowerJazz has a dedicated team with the proven technology and design enablement capabilities to meet our advanced process requirements for high performance, quality and technological support for our revolutionary product,” says Shay Kaplan, Audio Pixels Chief Scientist. ”In addition, their demonstrated commitment to provide expansive capacity and multi-fab sourcing makes TowerJazz an ideal partner to augment our future mass production requirements. ...”

"... TowerJazz provides world-class “Foundry MEMS Enablement” solutions which combine rapid implementation and ramp-up high volume production of both 150mm and 200mm MEMS manufacturing capabilities. “Audio Pixels’ MEMS-based loudspeaker is expected to have a significant impact on the multitude of industries and applications that currently consume billions of analog loudspeakers,” says Nachi Vofsy, Director of R&D at TowerJazz. “We are extremely pleased to be Audio Pixels’ foundry of choice for MEMS manufacturing to address these vast market opportunities.”

About TowerJazz
Tower Semiconductor Ltd. (NASDAQ: TSEM, TASE: TSEM) and its fully owned U.S. subsidiaries Jazz Semiconductor, Inc. and TowerJazz Texas Inc., operate collectively under the brand name TowerJazz, the global specialty foundry leader. TowerJazz manufactures integrated circuits, offering a broad range of customizable process technologies including: SiGe, BiCMOS, mixed-signal/CMOS, RF CMOS, CMOS image sensor, integrated power management (BCD and 700V), and MEMS. TowerJazz also provides a world-class design enablement platform for a quick and accurate design cycle as well as Transfer Optimization and development Process Services (TOPS) to IDMs and fabless companies that need to expand capacity.
To provide multi-fab sourcing and extended capacity for its customers, TowerJazz operates two manufacturing facilities in Israel (150mm and 200mm), two in the U.S. (200mm) and three additional facilities in Japan (two 200mm and one 300mm) through TowerJazz Panasonic Semiconductor Co. (TPSCo), established with Panasonic Corporation of which TowerJazz has the majority holding. Through TPSCo, TowerJazz provides leading edge 45nm CMOS, 65nm RF CMOS and 65nm 1.12um pixel technologies, including the most advanced image sensor technologies. For more information, please visit www.towerjazz.com or www.tpsemico.com. ..."

To refer back to Danny Lewin's address of May 2016 we are told that despite doing the initial development in a research FAB, for bulk manufacturing AP needed a close collaboration with a single MEMS production FAB.  This was later revealed to be TowerJazz.  If I can interpret this to mean that SONY was quite likely to have been the research FAB indicated below, then this gives no reason to believe that SONY was also given the job of mass producing bulk product in parallel to TJ.  If that had been the case why was this not also announced at the time of the TJ partnership? Why would the highly singular process of development be worked on in two independent FABs simultaneously?  Surely that would be prohibitively expensive, and would not necessarily produce identical device characteristics.

"... The effort to build a MEMS chip requires very close collaboration with MEMS foundries as unlike most conventional semiconductors which can take a chip design and apply fabricate processes with exceptional precision, the fabrication of any MEMS devices require the development of a specific and dedicated process flow for that particular device.
MEMS devices are produced in semiconductor fabs, using well-known and proven semiconductor equipment and processes. Although MEMS development can be done in research facilities (which is where we conducted our early research and development) we elected to conduct the commercialization plan in a mass production MEMS fab. Since mass production fabs are not oriented to accommodate the flexibility required in development, timelines tend to take a little longer. On the flip side this approach dramatically streamlines and accelerates the transition to mass production.
Because we don’t own our own MEMS fab, we are what is known as a fabless company which establishes a tight dependency on the fabrication partner. On the flip side we didn't have to hit shareholders up for many hundreds of millions of dollars required to build and run a fab. The high cost of setting up and running a MEMS fab dictates that Fab’s must carefully select which companies they work with, which projects they undertake and the priority of their resources. ..."

"... Of course all this get resolved in mass production phase when long term planning methodologies are implemented but in the interim the fabrication cycles are slow and occasionally fluctuate.
An additional challenge associated with MEMS development is that virtually any change to the device, no matter how minute, alters the mechanical behavior of the device. This can be critical as a few nanometers can change electrical response of sound pressure levels. While each process in of itself is predictable the integration of processes such as deposition, annealing, etching, coating etc.. can produce somewhat unpredictable results.
This reality necessitates that MEMS be developed using the trial and error methodology - measuring, dissecting, re-designing and re-fabricating over and over until the proceese as a whole is fine tuned to meet specific objectives. ...

... We are about two-thirds through a fabrication run with a particular fab that is using a fabrication process flow that has been tried, tested, and refined.
We perpetually monitor and test the results as for example is evident by the recent published results whereby we cycled 4th phase test structures over a trillion times – without a single failure.
Additionally we implemented a risk mitigating approach to the fourth phase - for example we decided to independently fabricate 8 separate batches of wafers, and each batch of wafers holds a number of wafers in reserve before undergoing critical fabrications steps as to allow for the possibility of taking corrective measures should it be deemed necessary.

The last comment above is what suggests to me that AP allowed itself plenty of room to move should  problems be encountered - which they were.  These comments also clearly suggest that a single particular FAB would have produced and would be expected to process all of the wafers.  

Notwithstanding this my conclusion would be that TJ has had more than one of its FAB lines set up to produce/process the wafers, and that the processes are transferrable between sites and lines, and that this has allowed internal second sourcing.  

One would presume (hope!) that the production design process would ensure that a first small batch of wafers was taken right through the process to the final stage, including MEMS production, integration with ASIC, and final packaging, before all of the intermediate wafers were committed to any further processing.  This would allow for the problems that we have been told about to be rectified on the bulk of the production run if required.  

One would hope that AP has been sufficiently prudent as to design its production process in such a way that it produces an initial batch of finished and validated devices for demonstration/evaluation purposes before committing all of its wafers to irrevocable processing.  The worst case scenario in this would be that the majority of the wafers have already been processed beyond all of the crucial stages before any flaws were discovered.  However the announcements to date do not indicate that this has been the case.  The fact that two parallel FABs or FAB lines are currently scheduled to deliver suitably modified production runs by 31/5/17 suggests that this was not the case.  It does not say that those two batches represent every wafer that has been produced.  Indeed it will not be necessary for any to be produced until orders are received by the company, and that is surely some time away yet.

I think (wishfully!) that we will get some reassuring outcomes from these runs sometime before 31/5/17.