Mesomorph machine architecture

The Mesomorph machine configuration comprises three stations

Femtosecond and 2PP unit

The femtosecond unit will present 5 DoFs organized in 3 linear mechanical axes and 2 optical ones. The mechanical configuration will present an overall production volume of 500x500x250 mm, thus enabling therefore a comprehensive batch manufacturing (Figure 2). The batch manufacturing will be ensured by an advance configuration of the optical chain where the high-power beam (500 W) will be split in multiple parallel beams that will in turn ensure the processing of multiple components on the same processing plate.

The current processing unit will be equipped with a laser head ensuring – through the same optical chain – also two photons polymerization 2PP. This technology ensures the fabrication of nearly arbitrary 3D micro- and macrostructures. By adjusting and controlling the photon dose only slightly beyond the polymerization threshold, 2PP ensures the realization of 100 nm scale features with extreme accuracy. 

Selective Area Direct Atomic Layer Processing unit

The SADALP unit will be configured with 3 DoF Cartesian architecture where the 2 planar movements will be located on the machine solid

basement structure and the vertical one will be directly on the vertical head; ensuring the same, 500x500x250mm, working cube of the previous module. In order to fulfil the strict planarity constraints (orthogonality between plane and nozzles) demanded by the process, an extremely accurate rototilt system is interposed between the head and the machine structure.

The scalability of such process to high productivity targets is ensured by the adoption of many independent nozzle-heads that will ensure the full coverage of the working area while enabling parallel processing of multiple components of the batch.

Micro-assembly and bonding unit

Between the two processing units mentioned above there is a third unit enabling handling and assembling capabilities to the machine. This unit is indeed designed to manipulate micro components such as silicon dies, sensors, opto-components, MEMS and more. The unit will automatically pick dies from Gel-Packs, Waffle Packs and Wafer Film frames onto substrates and packages. For components like MEMS, sensors and opto-devices this unit is provided with a special pick-up tool, supported by enhanced vision system, in order to avoid contact and potential damages to sensitive active areas. 

Micro-assembling within the machine is performed in the most reliable and effective way thanks to a multi-level neural network system which – in order to obtain the most robust strategies to precisely grasp and release parts – adopts the following strategy. In the first step, the training, the tiny components to be manipulated with all their physical specs are fed to a physics simulator which contains all the physical, static and dynamic information about the machine’s micro-gripper

The machine architecture is explicitly designed in order to handle processes involving hazardous and unhealthy environments. All the processing units will be engineered to operate in inert environments (e.g. Argon), where the oxygen percentage is kept below 0.5% (locally below 0.1%) and massive debris and fumes are generated. In order to isolate such processes and, at the same time, maximize the process productivity the two working areas are equipped with a set of sliding doors capable to completely separate the working areas from the micro-assembly one.

Mesomorph machine sensing system

Along with traditional process sensors, the sensing system will be composed of sensors operating in-line while the process is running (i.e. femto-laser monitoring camera, multi-camera system and imaging ellipsometer) and sensors operating in the machine but off-line with regard to the process (high speed interferometer). 

Schermata 2021-02-22 alle 10.27.03
Mesomorph Design/Manufacturing-to-Lifevalue digital Platform 

The Mesomorph digital infrastructure is conceived to cover the entire product lifecycle while focusing on the Design/Manufacturing-to-Lifevalue objective, from the very first idea of the customer to the realization and functional assessment of the finished part. It will be conceived and developed by MORPHICA and will include two components: 

Design-to-Lifevalue covers the design of complex micro-systems where the engineering stage accounts for the existence of multiple-technologies that will drastically transform the component shapes and functions. 

Manufacturing-to-Lifevalue covers the process engineering on the Mesomorph machine along with the inline product optimization triggered by the aforementioned sensing system.

Mesomorph Process Control

The advanced sensing system integrated in the Mesomorph machine enables the data capturing and fusion, which is the very premise to build an industrially robust process. In fact, the complexity of the manipulation tasks together with the challenges to engineer reliable process recipes for deposition and subtraction technologies demand for a deep activity in monitoring and controlling the process. Mesomorph Control will bringing innovation on the following aspects: right-the-first-time parts (NO SCRAPS); improved efficiency (energy and material saving); portability, i.e. solution initially developed for a specific machine setup, but, at the same time, scalable to the entire shop-floor.

The control solution will be developed as an adaptive hybrid process control and implemented as a control module called Mesomorph Adaptive Controller that will consider the entire process chain encompassing all the available technologies. Its ability to evolve the deposition/subtraction/manipulation behaviour relies upon a set of additional modules that will run off-line and on-line, over different time scales: Mesomorph Control Data Set Builder generates a process data set for creating a data history of parameters (this module will be nested to the KBS of the Design/Manufacturing-to-Lifevalue Platform); Mesomorph Control Empirical Modeller is an off-line module that, using the aforementioned data set, processes sensor and control data along with the electro-mechanical, geometrical and quality performance of the manufactured parts (optimal requirements) to generate a data driven process model by machine learning techniques;

Mesomorph Control Online Interpreter is an on-line sensor-based elaboration module that interprets the measurements captured during the deposition and identifies the in-control and out-of-control KPIs along with their trends. This will ensure Mesomorph Adaptive Controller to persistently close the loop on the:

  • laser power, spot, 2PP feeding dynamics and kinematic parameters on the femto/2PP unit;
  • reactants and carrier gases, stages dynamics on the SADALP unit;
  • strategies to grasp micro-components and strategies to adapt the path planning on the pick&place unit.

 

Machine learning will constitute the core approach for the Mesomorph Control Empirical model: given some inputs (e.g. manipulation parameters or real-time feedbacks from the laser beam), we will develop models with the goal of estimating one or more continuous variables (the set of quality parameters of the manufacturing workpieces) from known inputs, with the best possible accuracy. In detail, Deep Neural Networks – whose key feature stays in capturing complex patterns from huge training data sets in a variety of classification tasks – will be adopted for regression tasks; NN will act as universal, derivable approximators of an unknown mapping (thus integrating prior knowledge with the goal of process control). Additionally, NN will ensure: 

1) Through convolutional layers, high dimensional 2D-3D inputs can be managed; 

2) A derivable mapping between inputs and outputs, which comes instrumental for control purposes.