The development of the EuPRAXIA laser driver relies on Chirped Pulse Amplification (CPA) in Ti:Sapphire (Ti:Sa). The achievement of the required performance and reliability levels poses stark challenges related to several key elements in the system, and requires the most advanced components under development at the industrial level, along with specially devised solutions.
The implementation of the required plasma acceleration schemes involves the simultaneous use of up to three different laser systems, and the time of arrival of their output pulses on the plasma target must be carefully synchronized down to below 10 fs. Moreover, auxiliary laser beams for diagnostics and a photocatode laser are included, tightly synchronized with the main laser pulses.
In our study, the most critical points in the of design laser systems turned out to be the following:
Our design shows that the less challenging P0 performance level can be implemented with currently available technologies, requiring mainly integration efforts. Conversely, level P1 lies one step ahead with respect to the current technological capabilities, in particular regarding the PRR of the pump sources. Indeed, the choice of 100 Hz PRR was found to be the trade-off between the need for a sufficiently high repetition rate for user operation of the facility and the high technology readiness level (TRL) of the required components, assuming a 5-year time to construction.
2.2. Power Amplification
Pulse amplification relies on CPA in Ti:Sa, requiring pump sources in the visible range (i.e., frequency doubled, solid state Nd, or Yb-based lasers with emission in the range 515–532 nm). Several devices have been realized, achieving amplification of ultrashort (<20 fs) laser pulses with energy output of up to 100 J or greater, but these devices operate at a very low PRR (<1 Hz, or single shot). EuPRAXIA requires a much higher PRR, resulting in high average pump power requirements and a severe thermal load on the amplifiers.
Thermal load issues were addressed by means of water cooling at near room temperature of the end surfaces of the amplification crystals, shaped as disks with a relatively large diameter to thickness ratio, as recently proposed [
13]. Alternative approaches were considered, such as the cooling of the crystals by means of a high speed gas flow at cryogenic temperatures, as implemented in the DIPOLE Yb:YAG high energy laser system [
14,
15], or more recently in the Ti:Sapphire high energy amplifier implemented in the ELI-HAPLS system [
16]. This cooling method was nonetheless considered not suitable for this design, as it cannot provide a sufficient heat removal for this application, and it can hardly be scaled up to even higher thermal loads, as will be clarified in the following parts. For this reason, this solution was not further studied in the conceptual design.
The optimization of the extraction efficiency was carefully considered to reduce both the requirements on the pump lasers and the overall thermal load. This was addressed by a careful repartition of the amplification between the various stages, and by implementing the Extraction During Pumping (EDP) method to limit the buildup of transverse gain, and thus transverse parasitic lasing [
17,
18].
Modularity and scalability were addressed in the design—the various laser chains are built by different combinations of a limited set of amplification modules: Laser 1 (see
Figure 1) is composed of a single amplification stage (AMP1); Laser 2 consists of two amplification stages (AMP1, featuring the same design and operating parameters as in Laser 1, and AMP2); finally, Laser 3 consists of three stages (AMP1, AMP2, and AMP3), the first two being identical to Laser 2, and the third one featuring a dedicated design. The modules are already dimensioned to operate at the P1 level (
Table 1) if a sufficient pump pulse energy is available, as will be clarified in the following sections. This approach is advantageous in view of an industrial development of the system and in view of the scaling up of its performance from P0 to P1 along its lifetime.
2.3. Ti:Sapphire Amplifiers Structure and Geometry
A dedicated analysis was carried out to define the geometrical layout of the multi-pass amplifier based on thin disk crystals. To achieve sufficient pump absorption and energy storage, several disks (from 2 to 4) must be used in each amplification stage, depending on the pump energy and PRR. To achieve efficient amplification and energy extraction the amplified beam must cross the crystal several times (from 4 to 6 depending on the configuration).
As for the cooling strategy and amplification scheme, two possible approaches were considered for the disks: a transmission geometry and a reflection geometry.
In the transmission geometry, the disk-shaped crystal is cooled on both faces; both the pump beams and the amplified beam cross the crystal, the cooling water flow, and the flow containment windows. This solution (proposed in a previous study [
13] and analyzed in another study [
6] for the Eupraxia system) offers a good performance in terms of heat extraction and allows implementation of a simpler layout from a geometrical point of view, but it presents a potential drawback because the amplified beam is potentially subjected to optical perturbations due to the turbulence of the cooling fluid.
In the reflection geometry, one of the faces of the crystal is highly reflective for the amplification and pump beam. The amplified beam and the pump beams enter the crystal from the front face and are reflected back in the incoming direction on the cooled back surface (
Figure 2). As the laser beams do not cross the cooling flow, no optical perturbations are possible. This arrangement allows for less favorable surface/volume ratio for cooling and requires a more complex geometry.
Here, we describe in more detail the design of the amplifiers for the reflection geometry. Both design approaches are still under consideration, as it is difficult to assess the balance of the respective pros and cons only on the basis of theoretical considerations and simulations, and it will be addressed by suitable pilot studies.
The various laser modules were dimensioned by means of numerical simulations using the code MIRO developed by CEA (Commissariat à l’énergie atomique et aux énergies alternatives, France) [
12], validated by comparison with similar existing real laser systems [
18,
19]. As a baseline design, all of the amplifying modules feature a multi-pass amplification architecture with 4 passes. Here, we provide details of the amplifier AMP3 (see
Figure 1), as this is the amplifying unit running at the highest energy and power level, and thus is the most challenging in terms of thermal management. The unit consists of 3 equal amplification disks. The main parameters are reported in
Table 2 for the performance levels P0 and P1.
The geometrical layout of the disk arrangement is exemplified in
Figure 3, which shows separately the amplified beam path and the pump beam path. The seed pulse enters the first disk of the chain, and it is routed to the second and third disk by a suitable steering mirrors arrangement. After a first pass along the disk sequence, the pulse travels along a delay line and then it is sent backward along the chain meeting the three disks in reverse order. Angular multiplexing is used to separate the forward and the backward beam paths.
The three disks are pumped with the same amount of energy, equally distributed from the same pump pulse by beam-splitters (
Figure 3 right). On each disk, the pump energy is completely absorbed in four passes. The path of the pump beams is slanted in the vertical (z) plane to avoid obstruction by the steering mirrors for the amplified beam. The distance between the steering mirrors and the Ti:Sa disks is about 4 m, and the angles of the amplified beam path with respect to the disks’ surfaces are 1.2° and 2.4°.
The pump beam path includes suitable delay lines so that the arrival of the pump pulse on each disk is synchronized with the arrival of the amplified pulse.
The energy amplification process in AMP3 is described in
Figure 4, which reports the pulse energy for increasing numbers of passes, for the performance levels P0 and P1. For the P0 level the module is seeded with 18.8 J and pumped with 109 J, resulting in a output energy of 62 J; for the P1 performance level the output energy is 130 J with a seed pulse energy of 40 J, and a pump energy of 189 J.
The amplification process takes advantage of the fact that in both cases the seed energy is relatively high, so an efficient energy extraction can be realized even though the amplification per bounce on a single disk is relatively low. The different pulse duration and spectral bandwidth between P0 and P1 were found to have a negligible impact on the energy extraction.
The same concept was applied in the design of all the three modules (AMP1, AMP2, AMP3), as they operate at similar levels of pump pulse fluence, seed pulse fluence, and stored energy density. Therefore, the amplification of the pulse fluence is similar for all the three modules and it scales in the same way when going from P0 to P1 level. Energy scaling from AMP1 to AMP3 was mainly obtained by increasing the beam diameter at about constant fluence. In some cases a small adjustment in the beam diameter was needed to optimize the process either at P0 or at P1.
Regarding the thermal load, the various modules were designed to safely operate at the performance level P1; at the performance level P0 the thermal load is well below the design limits for each stage.
In conclusion, as was mentioned earlier, the output level of the amplifiers can be scaled up from P0 to P1 by increasing the pump and the seed energy without needing major modifications.
The occurrence of parasitic lasing was carefully analyzed in the design. Details about the methodology followed for this subject can be found in previous work [
6]. The choice of crystal diameter, beam size, crystal doping, and thickness result from an accurate tradeoff between longitudinal gain, energy storage, cooling considerations, and transverse lasing control and suppression. We notice here that although the multi-slab approach was primarily adopted to provide enough heat exchange surface area, as a side advantage it allows distribution of the stored energy in several crystals, reducing the transverse gain at the pump input face. Increasing the doping level would be favorable from the thermal point of view (as it would allow reduction of the crystal thickness for the same pump absorption), but it would result in an increased parasitic gain.
The implementation of the extraction during pumping (EDP) strategy relies on the proper synchronization between the injection of the seed pulse and of the pump pulse, and on the presence of the delay lines for the pump beam between the various disks.
On each disk, the pump pulse is almost completely absorbed in four passes, using the pump recycling mirrors PRM1, PRM2, and PRM3 to send back in the crystal the fraction transmitted after the first bounce.
The time required to complete the pump absorption process on each disk from the moment when the leading edge of the pump pulse first hits the disk depends on the pump pulse duration and on the distance between the disk and the PRM. As a numeric example, if the pump pulses are roughly rectangular with a duration of 15 ns, and the distance between the disks and the PRM is 2.5 mt, the pump absorption will be completed in about 30 ns on each disk.
The seed pulse arrives on the disk before the end of the absorption period, so that the available energy for amplification is only the fraction of the pump pulse absorbed before the arrival of the seed pulse itself; after the first bounce, the seed pulse leaves the disk and it returns back only for the return pass. During this interval the energy absorption is completed and the stored energy remains available for the amplification in the return pass, as the lifetime of the Ti:Sa is obviously much longer than the overall pulse transit time in the disks sequence.
This sequence of events can be replicated in each disk along the chain by keeping the arrival time of the pump pulse on each disk properly synchronized with the arrival of the seed pulse. For this purpose, the additional pump delay lines between each disk (mirrors PDM1 and PDM2) defer the arrival of the pump pulse on each disk exactly by the time that the seed pulse takes to complete its path through the previous Ti:Sa disks.
This timing scheme is exemplified in the diagram in
Figure 5. Through this approach, the stored energy repartition between the first and the second bounce on each disk can be finely tuned by adjusting the injection delay between the pump pulse and the seed pulse, and the length of the intermediate delay lines. As the pump absorption process on each disk takes a time in the order of several tens of ns, a timing accuracy in the range of ~0.3–0.5 ns can allow a control of the energy repartition with about 1% accuracy.
2.4. Fluid Cooling Simulations
A simulation of the temperature distribution and of the thermomechanical stresses affecting the amplifier disks was carried out by means of Finite Element Analysis (FEA) software available in a commercial software package, i.e., LAS-CAD (ver. 3.6.1) developed by LAS-CAD GmbH (
www.las-cad.com), to calculate the spatial temperature distribution in the gain material, the resulting stress and deformation distribution induced by the thermal expansion, and finally the thermal aberrations computing the optical path difference (OPD) distribution across the crystal aperture due to both the variation of the refractive index with temperature and the variation in the crystal thickness due to thermal expansion and thermally induced stresses. Thermally induced birefringence [
20] was not considered. Detailed results of this approach are described in previous work [
6].
A key point to obtain meaningful results from the FEA simulations is the modelling of the heat exchange between the solid and the fluid. This was modelled by means of a film coefficient k = Q/ΔT, where ΔT is the local difference of temperature between the solid surface and the fluid, and Q is the power per unit surface transferred from the solid to the fluid. Although approximate, this approach is very effective in reducing the computational effort to acceptable levels, in particular with full 3D models.
The fluid–solid heat transfer process was studied by means of dedicated fluidodinamical simulations, in 2D geometry, in order to obtain reliable values for the heat transfer coefficient used in the full 3D simulation. The computational method was a numerical solution of the Navier–Stokes equations for mass and energy transport, using the so-called Low Reynolds k-ε method, which is capable of a quite accurate simulation of the behavior of the fluid-solid interface layer with a reasonable computational effort. The fluidodynamics simulations were carried out using a commercial software package (COMSOL Multiphysics Version 5.3a).
An example is shown in
Figure 6, depicting the fluid flow in contact with the back (reflective) surface of the disk of
Figure 2. The simulation considers a flow of water at 6 m/s entrance speed, in a cooling channel with a thickness of 5 mm. The fluid flow is heated from its bottom side by a constant heat flux of 25 W/cm
2, corresponding to the heat input generated in the Ti:Sapphire disk under pumping at the P1 performance level. The length of the heated portion of the channel is 160 mm (corresponding to the diameter of the disk of AMP3).
Figure 5 shows the calculated fluid velocity distribution. The temperature dependence of the water properties (viscosity, density) was taken into account; buoyancy effects were considered negligible.
Figure 6 also shows the calculated temperature distribution of the boundary layer of the water flow in contact with the Ti:Sapphire surface for different flow speeds. Under fully developed heat transfer conditions the temperature increase of the surface is about proportional to the heat input. This allows calculation of the effective film coefficient
k =
Q/Δ
T, which expresses the cooling capability of the fluid flow. The effective film coefficient resulted independently from of the heat input, and proportional to the flow speed (proportionality constant of about 0.04 J/cm
3). At the maximum flow speed of 6 m/sec the film coefficient was 2.45 W/(cm
2 K).
With the channel configuration shown in
Figure 6 (i.e., parallel walls), only a very thin layer of fluid near the surface is actually involved in the heat exchange process. Other channel configurations were also studied, with ridges meant to enhance the fluid turbulence and increase the heat exchange, as shown in
Figure 7. Due to turbulence, the fluid speed locally increases (up to about twice the input velocity), thus improving the heat transfer from the lower surface. The temperature increase of the heated surface is thus lower by more than a factor of 2 with respect to the previous case, with a corresponding increase in the film coefficient
k.
Such a configuration can be applied for the cooling of the disks in the reflection configuration, as the back face of the cooling channel is not crossed by the light beams, so it can be shaped to optimize heat exchange processes.
It can be noticed that the transversal flow configuration induces a temperature gradient in the cooling flow (and thus in the cooled surface of the Ti:Sa disk) in the flow direction as the fluid is heated along its path. This induces some amount of thermal aberration that is not radially symmetric, as would be desirable. On the other hand, the overall amount of this wavefront distortion is relatively small, at least for the highest flow speeds considered here, as it can be easily estimated. From the graph of
Figure 7 it can be seen that the difference in the Ti:Sa surface temperature between the input edge and the output edge is about 1.6 °C. As the transverse heat flow in the Ti:Sa disk is negligible (due to the thin disk geometry), this temperature difference is directly added to the internal temperature distribution in the Ti:Sa. Accounting for the thermal expansion coefficient and for the value of dn/dT of the Ti:Sa (5 × 10
−5/K and 1.5 × 10
−5/K, respectively), the thermally induced optical path difference is about 0.4 µm over a single pass (i.e., about λ/2), with a roughly linear variation along the disk diameter. Moreover, this specific contribution to the wavefront aberration is stationary or at worst slowly varying in time, so it can be easily mitigated or canceled both by a proper geometrical arrangement (e.g., by having the flow propagating in alternate opposite directions in the disks) and by active wavefront correction optics, which are already considered in the design.
These investigations are to some extent preliminary, and a significantly broader work is required to obtain an executive design. Particularly relevant are the problems related to the vibrations induced by turbulences and by the pumping system, which could be detrimental for the beam pointing stability, in particular in the reflection geometry. Our major concern here was to assess the effectiveness of the water cooling approach at this unprecedented level of thermal load.
We notice here that according to the data reported in existing literature, gas cooling cannot provide a heat removal rate comparable with water cooling. Implementation of gas cooling would thus require unreasonably large heat exchange surfaces to obtain sufficient cooling, in particular at the envisaged P1 performance level. In more detail, currently the highest cooling performance obtained by this technique in the DiPOLE 100 system corresponds to the removal of about 4 kW of thermal load, with an available cooling surface of about 1450 cm
2 (six Yb:YAG slabs with a cross-section of 11 × 11 cm
2, cooled on both sides) [
15]; the average heat flow per unit cooling surface is less than 3W/cm
2, and the film coefficient
k was estimated to about 0.17 W(cm
2 K) [
14]. On the other hand, the overall thermal load expected in the AMP3 module is about 10 kW (i.e., a factor 2.5-times larger than in DiPOLE 100), with an average heat flow per unit cooling surface as high as 25 W/cm
2 (i.e., an order of magnitude higher); the value of the heat transfer coefficient obtained with water cooling (2.45 W/(cm
2 K) is more than an order of magnitude larger than in the case of He cooling. We notice here that cryogenic He cooling was also recently employed for the cooling of a Ti:Sa multi-slab amplifier for the HAPLS system [
16]; however, in that case the system runs at 3.3 Hz, with a pump energy of about 60 J, setting the average pump power to about 200 W. This is much lower than the average pump power expected in the Eupraxia system for any of the amplification stages.