清华大学 机械工程系, 摩擦学国家重点实验室, 精密超精密制造装备及控制北京市重点实验室, 北京 100084
收稿日期:2022-06-24
基金项目:国家科技重大专项(2017ZX02102004)
作者简介:刘涛(1996-), 男, 博士研究生
通讯作者:杨开明, 副研究员, E-mail: yangkm@tsinghua.edu.cn
摘要:前馈控制器是光刻机工件台在高加减速工况下实现纳米级运动精度的关键环节。该文针对传统情况下四阶前馈拟合逆模型能力较差、难以完全消除参考轨迹导致重复性误差的问题, 提出了一种以四阶前馈为基础, 外加有理分式补偿器的前馈控制架构, 并针对该有理分式补偿器控制参数整定的问题, 提出了一种数据驱动的参数整定方法。该方法利用系统辨识的相关规则, 将前馈补偿器参数整定过程的非凸优化问题转化为凸优化问题, 进而给出了全局最优参数整定方法以及参数迭代过程中梯度、Hessian矩阵的无偏估计方法; 通过光刻机工件台的实验验证了所提参数整定方法具有收敛性。实验结果表明:所提出的补偿前馈能够有效消除四阶前馈未消除的残余误差。
关键词:前馈控制有理前馈控制器补偿前馈数据驱动参数整定工件台
Parameter tuning of the wafer stage compensation feedforward controller of the lithography machine
LIU Tao, YANG Kaiming, ZHU Yu
Beijing Key Laboratory of Precision/Ultra-Precision Manufacturing Equipment and Control, State Key Laboratory of Tribology, Department of Mechanical Engineering, Tsinghua University, Beijing 100084, China
Abstract: [Objective] The feedforward controller is crucial to achieving nano-level motion accuracy for the lithography wafer stage under high acceleration and deceleration conditions. Traditional 4-order feedforward is widely used to control precision motion systems because of its intuitive physical meaning and simple parameter tuning. However, its capacity to fit the inverse model is inadequate, and it is difficult to eliminate the repetitive error caused by the input trajectory. Therefore, a feedforward control architecture using the 4-order feedforward and an extra rational fraction compensator is proposed. [Methods] In this study, the input signal of the compensator is the higher-order derivative of the reference trajectory, and the numerator and denominator of the compensator use the delay unit as the basis function. Therefore, obtaining the unknown parameters of the basis function is crucial to the design. This paper proposes a data-driven iterative parameter tuning strategy for the compensation controller. The difficulty is that the tuning problem is a nonconvex optimization problem, making global parameter optimization challenging. This paper uses the relevant rules of system identification to address the issue at hand. The purpose of adding compensatory feedforward is to eliminate the residual error after using the 4-order feedforward, which is equivalent to achieving a zero-generalized error. Since the generalized error has a linear connection with the compensator parameters, the original nonconvex optimization problem is successfully transformed into a convex problem by minimizing the 2-norm of the generalized error. Through the above transformation, the global optimal point is obtained by the Gauss—Newton method, and the step size condition for ensuring iterative convergence is provided. In addition, the gradient and Hessian matrix of the objective function need to be incorporated into the parameter updating law, even though their exact values are difficult to obtain. This paper derives their unbiased estimates using two impulse response experiments and 2 trajectory tracking experiments. [Results] The proposed method was applied to the wafer stage of the lithography machine, and the experiment showed the following results: (1) Using the proposed method to tune three compensation controllers with different orders, their error 2-norm almost converged after five iterations. (2) After adding compensation feedforward, the acceleration and deceleration phase errors were reduced from ±35 nm to ±10 nm; the constant velocity phase error was almost equal to the positioning error, and its trajectory tracking effect was very close to that of iterative learning control (ILC) compensation. (3) Compared with the existing compensation controller parameter tuning method, the maximum moving average and moving standard deviation at velocity phase of the proposed method were smaller, and the lower the compensator order, the more obvious the advantage. (4) After changing trajectory, the proposed compensator could still achieve a better control effect than ILC compensation. [Conclusions] The above experiments verify the convergence performance of the proposed parameter tuning algorithm. It is shown that the proposed feedforward compensation architecture can effectively eliminate the residual repetition error of the 4-order feedforward; simultaneously, it can adapt to variable trajectories. In addition, compared to the current compensator tuning result, this method can achieve a superior trajectory tracking control effect while using a low-order compensation controller.
Key words: feedforward controllerrational feedforward controllercompensation feedforwarddata-drivenparameter tuningwafer stage
以光刻机工件台为典型代表的超精密运动系统对高速、高加减速工况下的运动精度有极高要求,前馈控制技术是实现超精密运动系统严苛的轨迹跟踪精度目标的重要手段[1-3]。
前馈控制的基本思路是拟合被控对象的逆模型,其中比较经典的方法是基于模型(model-based)的前馈[4-6]。然而,光刻机工件台一般都具有结构复杂、模型不确定性大、柔性振动模态凸显等特征,难以获得精确的模型结构及参数,不可避免地会产生建模误差;基于模型的前馈对系统非最小相位零点的近似处理也会产生求逆误差;上述建模、求逆误差的存在导致基于模型的前馈难以完全拟合被控对象模型的逆,最终难以达到光刻机工件台纳米级运动精度的要求[7]。
相比之下,数据驱动(data-driven)的方法避免了建模、求逆的过程,将系统视作黑箱,通过直接采集系统的输入、输出等控制过程数据,将控制器的参数整定问题转化为关于某一目标函数的优化问题[3, 8]。根据前馈形式的不同,数据驱动前馈可进一步分为基于信号(signal-based)的前馈和基于控制器结构(structure-based)的前馈。
迭代学习控制(iterative learning control, ILC)是一类典型的基于信号的前馈[9-10]。对于重复性轨迹跟踪问题,ILC通过利用上次实验的误差信号生成下次实验的前馈补偿量,理论上可以完全消除重复性误差,是轨迹跟踪精度最高的控制算法之一。然而ILC严重依赖初始条件,当输入的参考轨迹发生变化时,控制效果可能会严重变差,此时需要重新学习补偿信号[11]。
基于控制器结构的前馈具有更强的轨迹泛化性能,基本思路与基于模型的前馈一致,但不再利用模型相关信息来逼近逆模型,而是通过优化一组与模型无关的基函数的系数来逼近逆模型[12-13]。最常用的是多项式类型前馈,如四阶前馈[14],采用微分算子作为基函数,将参考轨迹的速度(velocity)、加速度(acceleration)、加速度一阶导数(jerk)、加速度二阶导数(snap)信号的加权组合作为前馈补偿信号,由于具有物理直观性,该方法被广泛应用在光刻机工件台[15]、打印机[14]等精密运动系统的控制中。多项式类型前馈的线性参数化结构保证了前馈控制器一定是稳定的,并且控制器参数整定为凸优化问题也易于获取最优参数,但该线性参数化结构也导致只能拟合含有分母的被控对象,难以补偿系统的高频段柔性谐振模态。针对上述问题,文[16-17]提出了有理分式类型前馈。相比于多项式的有限冲击响应结构,有理分式的无限冲击响应结构使有理分式类型前馈具备更强的逆模型拟合能力,但同时也导致前馈控制器参数整定为非凸问题,一般的参数优化算法难以保证获取最优值[12],并且在迭代过程中还需要考虑前馈控制器分母的稳定性问题。
综上所述,无论是基于模型的前馈,还是基于控制器结构的数据驱动前馈,拟合逆模型的精度受限于前馈选择的形式,最终会不可避免地产生拟合残差;在关注的频段范围内,该拟合残差会严重影响系统最终的运动精度。因此需要在上述前馈基础上添加前馈补偿项,进一步提高轨迹跟踪精度[18-19]。由于ILC训练得到的前馈控制信号中包含丰富的模型信息,有****提出了通过拟合ILC信号来整定补偿前馈控制器参数的方法[18, 20-21]。
为减少逆模型的拟合残差对光刻机工件台控制性能的影响,本文在现有四阶前馈的基础上,提出一种新的通过数据驱动方法迭代整定参数的前馈补偿结构。通过优化广义残余重复误差,将有理分式结构补偿前馈参数整定过程中的非凸优化问题转化为凸优化问题,进而基于Gauss-Newton法给出了补偿前馈全局最优参数更新率,同时给出了参数整定过程中梯度、Hessian矩阵及有关变量的无偏估计方法。
1 光刻机工件台及其控制方法1.1 光刻机工件台在集成电路的制造过程中,光刻机工件台用于承载晶圆,完成测量、曝光等工艺流程。清华大学机械工程系IC装备实验室开发的一款光刻机工件台如图 1所示,采用粗精叠层结构:粗动台由直线电机驱动,该直线电机的磁钢固定在大理石台上,动子线圈固定在粗动台上,通过直线导轨使粗动台在Y向进行大行程微米级运动;微动台叠加在粗动台上,由4个水平音圈电机和4个垂向音圈电机产生六自由度运动,通过固定在机架上的亚纳米分辨率激光干涉仪反馈实时位移,实现小行程纳米级运动。光刻机高产率(≥295硅片/h)、高套刻精度(≤2.5 nm)的整机指标需要工件台在高速、高加速的运动条件下实现纳米级的运动精度,为控制带来极大的挑战。
图 1 光刻机工件台 |
图选项 |
微动台运动系统通常可由多质量块模型建模,其传递函数一般可表示为[18-19]
$P(s)=\frac{1}{M s^{2}}+\frac{1}{M} \sum\limits_{i=1}^{I} \frac{\gamma_{i}}{s^{2}+2 \eta_{i} \omega_{i} s+\omega_{i}^{2}} .$ | (1) |
1.2 工件台控制方法工件台运动系统通常采用“前馈+反馈”的二自由度控制结构[2, 8],如图 2所示。
图 2 二自由度控制结构 |
图选项 |
图 2中:
当不考虑
$e=\underbrace{S(s) r-S_{\mathrm{p}}(s) F r}_{e_{F}}-\underbrace{S(s) v}_{e_{v}} .$ | (2) |
$F(s)=P^{-1}(s) \text {. }$ | (3) |
由于光刻机工件台微动台的无阻尼特性,通常
$F(s)=\varepsilon_{\text {acc }} s^{2}+\varepsilon_{\text {jerk }} s^{3}+\varepsilon_{\text {snap }} s^{4} .$ | (4) |
$P^{-1}(s)=\mu_{1} s^{2}+\mu_{2} s^{3}+\mu_{3} s^{4}+\Delta G(s) .$ | (5) |
$e_{F}=S_{\mathrm{p}} \cdot \Delta G \cdot r .$ | (6) |
1.3 额外补偿前馈额外补偿力信号可通过以下2种方式获取。
1) 基于信号的补偿控制。
通过ILC获取补偿力信号, 即
$u_{\mathrm{ILC}}^{k+1}=L_{1}\left(u_{\mathrm{ILC}}^{k}+L_{2} e^{k}\right) \text {. }$ | (7) |
2) 基于控制器的补偿控制。
通过额外补偿前馈控制器
$u_{\mathrm{c}}=\Delta F \cdot r^{p} .$ | (8) |
$\begin{gathered}\Delta F=z^{d} \cdot \frac{B\left(z^{-1}, \boldsymbol{\theta}_{B}\right)}{A\left(z^{-1}, \boldsymbol{\theta}_{A}\right)}= \\z^{d} \cdot \frac{b_{0}+b_{1} z^{-1}+\cdots+b_{m} z^{-m}}{1+a_{1} z^{-1}+\cdots+a_{n} z^{-n}} .\end{gathered}$ | (9) |
图 3 基本前馈、额外补偿前馈控制结构 |
图选项 |
由于基于信号的补偿前馈的变轨迹适应能力较差,故而通常采用基于控制器结构的补偿前馈。此时,问题的核心变为如何整定
2 数据驱动额外补偿前馈参数整定方法假设
$e(\boldsymbol{\theta})=\underbrace{S r-S_{\mathrm{p}} F r}_{e_{F}}-\frac{B\left(z^{-1}, \boldsymbol{\theta}_{B}\right)}{A\left(z^{-1}, \boldsymbol{\theta}_{A}\right)} S_{\mathrm{p}} \cdot z^{d} \cdot r^{p}-S v$ | (10) |
2.1 全局最优参数整定方法借鉴文[12]和系统辨识的相关方法,可将上述非凸优化问题转化为凸优化问题。设计ΔF的目的是补偿基本四阶前馈没有消除的残余重复性误差,使下式成立:
$e_{F}-\frac{B\left(z^{-1}, \boldsymbol{\theta}_{B}\right)}{A\left(z^{-1}, \boldsymbol{\theta}_{A}\right)} S_{\mathrm{p}} z^{d} r^{p}=0 .$ | (11) |
$\eta_{r}(\boldsymbol{\theta})=A\left(z^{-1}, \boldsymbol{\theta}_{A}\right) e_{F}-B\left(z^{-1}, \boldsymbol{\theta}_{B}\right) S_{\mathrm{p}} z^{d} r^{p} .$ | (12) |
$\begin{gathered}\eta_{r}(\boldsymbol{\theta})=e_{F}+z^{-1} \boldsymbol{\phi}_{n-1}(z) e_{F} \boldsymbol{\theta}_{A}- \\\boldsymbol{\phi}_{m}(z) S_{\mathrm{p}} z^{d} r^{p} \boldsymbol{\theta}_{B}=e_{F}-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }} \boldsymbol{\theta} .\end{gathered}$ | (13) |
$\left\{\begin{array}{l}\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}=\left[\begin{array}{ll}-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{A} & \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{B}\end{array}\right], \\\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{A}=z^{-1} \boldsymbol{\phi}_{n-1}(z) e_{F}, \\\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{B}=\boldsymbol{\phi}_{m}(z) S_{\mathrm{p}} z^{d} r^{p} .\end{array}\right.$ | (14) |
$J(\boldsymbol{\theta})=\frac{1}{2}\left\|\eta_{r}(\boldsymbol{\theta})\right\|_{Q}^{2}=\frac{1}{2}\left(\eta_{r}(\boldsymbol{\theta})\right)^{\mathrm{T}} \boldsymbol{Q}_{\eta_{r}}(\boldsymbol{\theta}) .$ | (15) |
$\boldsymbol{\theta}^{k+1}=\boldsymbol{\theta}^{k}-\alpha^{k}\left(\nabla^{2} J\left(\boldsymbol{\theta}^{k}\right)\right)^{-1} \nabla J\left(\boldsymbol{\theta}^{k}\right) .$ | (16) |
$\left\{\begin{array}{l}\nabla J\left(\boldsymbol{\theta}^{k}\right)=(-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }})^{\mathrm{T}} \boldsymbol{Q}_{\eta_{r}}\left(\boldsymbol{\theta}^{k}\right), \\\nabla^{2} J\left(\boldsymbol{\theta}^{k}\right)=\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}^{\mathrm{T}} \boldsymbol{Q} \mathit{\boldsymbol{ \boldsymbol{\varPhi} }} .\end{array}\right.$ | (17) |
1)
$\hat{e}_{F}=S r-S_{\mathrm{p}} F r-S v .$ | (18) |
2) 由于
$\begin{gathered}\hat{\eta}_{r}\left(\boldsymbol{\theta}^{k}\right)=A\left(z^{-1}, \boldsymbol{\theta}_{A}^{k}\right) e\left(\boldsymbol{\theta}^{k}\right)= \\\eta_{r}\left(\boldsymbol{\theta}^{k}\right)-A\left(z^{-1}, \boldsymbol{\theta}_{A}^{k}\right) S v .\end{gathered}$ | (19) |
3)
实际计算中需要将离散域形式转化为代数形式,此时离散传递函数变为其单位脉冲响应信号对应的Toeplitz矩阵[1]。后文以相同符号的粗体表示传递函数对应的Toeplitz矩阵。
采用图 4所示脉冲响应(impulse response)实验即可无偏估计
图 4 脉冲响应实验估计Sp |
图选项 |
给定参考轨迹
$\underbrace{\left[\begin{array}{c}y(0) \\y(1) \\\vdots \\y(N-1)\end{array}\right]}_{\boldsymbol{y}}=\boldsymbol{S}_{\mathrm{p}} \underbrace{\left[\begin{array}{c}1 \\0 \\\vdots \\0\end{array}\right]}_{\boldsymbol{u}_{\text {imp }}}+\boldsymbol{S}\left[\begin{array}{c}v(0) \\v(1) \\\vdots \\v(N-1)\end{array}\right] .$ | (20) |
$\begin{aligned}& \underbrace{\left[\begin{array}{cccc}y(0) & 0 & \cdots & 0 \\y(1) & y(0) & \cdots & 0 \\\vdots & \vdots & & \vdots \\y(N-1) & y(N-2) & \cdots & y(0)\end{array}\right]}_{\hat{s}_{\mathrm{p}}}=\boldsymbol{S}_{\mathrm{p}}+ \\& \boldsymbol{S}\underbrace{\left[\begin{array}{cccc}v(0) & 0 & \cdots & 0 \\\mid v(1) & v(0) & \cdots & 0 \\\vdots & \vdots & & \vdots \\v(N-1) & v(N-2) & \cdots & v(0)\end{array}\right]}_{\boldsymbol{v}} . \end{aligned}$ | (21) |
获取上述3个中间变量的无偏估计之后,得出如下结论:
定理 ??通过2次基本四阶前馈轨迹跟踪实验和2次脉冲响应实验, 即可获得
证明 ??2次基本四阶前馈轨迹跟踪实验
$\left\{\begin{array}{l}\hat{e}_{F, \mathrm{H} 1}=e_{\mathrm{H} 1}=e_{F}-S v_{\mathrm{H} 1}, \\\hat{e}_{F, \mathrm{H} 2}=e_{\mathrm{H} 2}=e_{F}-S v_{\mathrm{H} 2}, \\\hat{\boldsymbol{S}}_{\mathrm{p}, \mathrm{H} 3}=\boldsymbol{S}_{\mathrm{p}}+\boldsymbol{S} \boldsymbol{V}_{\mathrm{H} 3}, \\\hat{\boldsymbol{S}}_{\mathrm{p}, \mathrm{H} 4}=\boldsymbol{S}_{\mathrm{p}}+\boldsymbol{S} \boldsymbol{V}_{\mathrm{H} 4} .\end{array}\right.$ | (22) |
$\left\{\begin{array}{l} \mathit{\boldsymbol{ \boldsymbol {\hat\varPhi} }}_{1}=\left[-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{A, \mathrm{H} 1} \quad \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{B, \mathrm{H} 3}\right]=\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}+ \\ \underbrace{\left[-z^{-1} \boldsymbol{\phi}_{n-1}(z) S v_{\mathrm{H} 1} \boldsymbol{\phi}_{m}(z) \boldsymbol{S} \boldsymbol{V}_{\mathrm{H} 3} z^{d} r^{p}\right]}_{\Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{1}}, \\ \mathit{\boldsymbol{ \boldsymbol {\hat\varPhi} }}_{2}=\left[\begin{array}{ll}-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{A}, \mathrm{H} 2 & \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{B}, \mathrm{H} 4\end{array}\right]=\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}+ \\ \underbrace{\left[-z^{-1} \boldsymbol{\phi}_{n-1}(z) S v_{\mathrm{H} 2} \quad \boldsymbol{\phi}_{m}(z) \boldsymbol{S} \boldsymbol{V}_{\mathrm{H} 4} z^{d} r^{p}\right]}_{\Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{2}} .\end{array}\right.$ | (23) |
$\begin{gathered}E\left\{\hat{\nabla}^{2} J\left(\boldsymbol{\theta}^{k}\right)\right\}=E\left\{\hat{\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}}_{1}^{\mathrm{T}} \boldsymbol{Q} \hat{\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}}_{2}\right\}= \\E\left\{\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}^{\mathrm{T}} \boldsymbol{Q} \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}+\left(\Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{1}\right)^{\mathrm{T}} \boldsymbol{Q} \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}+\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}^{\mathrm{T}} \boldsymbol{Q} \Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{2}+\right. \\\left.\Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{1}^{\mathrm{T}} \boldsymbol{Q} \Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_{2}\right\}=\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}^{\mathrm{T}} \boldsymbol{Q} \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}=\nabla^{2} J\left(\boldsymbol{\theta}^{k}\right) .\end{gathered}$ | (24) |
$\begin{gathered}E\left\{\hat{\nabla} J\left(\boldsymbol{\theta}^k\right)\right\}=E\left\{-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_1^{\mathrm{T}} \boldsymbol{Q} \hat{\eta}_r\left(\boldsymbol{\theta}^k\right)\right\}=-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}^{\mathrm{T}} \boldsymbol{Q} \eta_r\left(\boldsymbol{\theta}^k\right)- \\E\left\{\Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_1^{\mathrm{T}} \boldsymbol{Q} \eta_r\left(\boldsymbol{\theta}^k\right)-\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}^{\mathrm{T}} \boldsymbol{Q} A\left(z^{-1}, \boldsymbol{\theta}_A^k\right) S v-\right.\\\left.\Delta \mathit{\boldsymbol{ \boldsymbol{\varPhi} }}_1^{\mathrm{T}} \boldsymbol{Q} A\left(z^{-1}, \boldsymbol{\theta}_A^k\right) S v\right\} .\end{gathered}$ | (25) |
2.3 参数收敛性能及额外补偿前馈控制器的稳定性理想的最优(optimal) 补偿前馈参数
$\eta_{r}(\boldsymbol{\theta})=\mathit{\boldsymbol{ \boldsymbol{\varPhi} }}\left(\boldsymbol{\theta}_{\mathrm{opt}}-\boldsymbol{\theta}\right) .$ | (26) |
$\boldsymbol{\theta}^{k+1}-\boldsymbol{\theta}_{\mathrm{opt}}=\left(1-\alpha^{k}\right)\left(\boldsymbol{\theta}^{k}-\boldsymbol{\theta}_{\mathrm{opt}}\right) \text {. }$ | (27) |
此外, 在参数整定过程中, 有理分式结构中分母
3 实验结果与分析3.1 实验设置本文以图 1中光刻机工件台的微动台为实验对象。由于预先解耦降低了各自由度间的影响,其Y向可视为单自由度问题,因此本文只关注Y向的轨迹跟踪误差。
反馈控制器由经典的比例-积分-微分(proportional-integral-derivative, PID)控制器、陷波滤波器构成[8]。通过调节反馈控制器,使工件台定位误差为
1) M1: ? 基于数据驱动的
2) M2: ?文[18]方法,首先进行迭代学习补偿实验,通过式(7)获得
关于何时停止迭代, 可通过定义收敛准则进行判定, 可参考文[19]; 但是对于ILC或凸优化问题, 收玫速度很快, 在本文实验中,
3.2 轨迹跟踪实验结果光刻机工件台的参考轨迹常用四阶轨迹[1, 12],本文实验中所使用的四阶轨迹
表 1 轨迹R1和R2的参数信息
指标 | R1 | R2 |
行程/mm | 300 | 300 |
扫描速度/(m·s-1) | 2 | 2 |
最大加速度/(m·s-2) | 50 | 52 |
最大加速度一阶导数/(m·s-3) | 3 200 | 5 000 |
最大加速度二阶导数/(m·s-4) | 320 000 | 640 000 |
表选项
本文实验中, 选择
图 5 误差2范数的平方随迭代次数的变化曲线 |
图选项 |
图 6 轨迹跟踪误差曲线对比 |
图选项 |
在基本四阶前馈的基础上,分别添加的ILC补偿、M1方法补偿,结果如图 5和6所示,可以看出:
1) 图 5中,M1方法3种阶次的
2) 图 5中,使用M1方法进行补偿时,误差2范数的平方由2×105下降至2×104,接近ILC补偿的水平;图 6中,添加
ILC得到的补偿信号通过M2方法同样可以得到3种阶次的
图 7 轨迹R1下M1、M2控制效果对比 |
图选项 |
1) 无论是对比误差2范数的平方, 还是对比误差的幅值, M1整定得到的
2) 对M1而言,图 7c和7d中轨迹跟踪误差的幅值与定位误差基本相当,可认为其主要呈现噪声特性,残余重复性误差分量基本被消除。
3) 对M2而言,当整定阶次较低时,通过M2整定得到的
因此,为了实现良好的补偿效果,M2通常会给出一个阶次较高的前馈控制器。但是对于工件台运动系统而言,其伺服周期较短,留给计算控制补偿量的时间有限,而阶次较高的前馈控制器会增加计算时间。
此外,匀速段是工件台用于曝光的阶段,也是其性能被严格要求的阶段。工件台具体的运动性能指标为移动平均值(moving average, MA)和移动标准差(moving standard deviation, MSD)[8],具体可表示为
$\left\{\begin{array}{l}M_{\mathrm{a}}(k)=\frac{1}{l} \sum\limits_{i=k-l / 2}^{k+l / 2-1} e(i), \\M_{\mathrm{sd}}(k)=\sqrt{\frac{1}{l} \sum\limits_{i=k-l / 2}^{k+l / 2-1}\left(e(i)-M_{\mathrm{a}}(k)\right)^{2}} .\end{array}\right.$ | (28) |
表 2 轨迹R1下匀速段误差的Ma、Msd值
类别 | ‖Ma‖∞/nm | ‖Msd‖∞/nm | |||
M1 | M2 | M1 | M2 | ||
五阶 | 3.46 | 5.07 | 7.27 | 14.16 | |
4.56 | 4.71 | 6.56 | 12.17 | ||
3.44 | 4.25 | 7.08 | 10.69 | ||
均值 | 3.82 | 4.67 | 6.97 | 12.34 | |
八阶 | 3.68 | 4.21 | 5.72 | 13.21 | |
4.20 | 3.51 | 6.50 | 10.67 | ||
3.68 | 4.33 | 5.90 | 11.70 | ||
均值 | 3.85 | 4.01 | 6.04 | 11.86 | |
十阶 | 3.69 | 4.60 | 6.32 | 8.04 | |
4.08 | 4.11 | 5.42 | 6.51 | ||
4.17 | 3.90 | 5.64 | 6.71 | ||
均值 | 3.98 | 4.20 | 5.79 | 7.09 |
表选项
由表 2可知,M1的控制效果明显优于M2。此外,随着
将
图 8 轨迹R2下M1、M2控制效果对比 |
图选项 |
3.3 结果分析前述实验验证了本文所提算法有效,本节就上述结果展开进一步的分析。
1) 补偿前馈
理想的补偿前馈可表示为
$\Delta F=P^{-1}-F \text {. }$ | (29) |
2) 基于ILC信号拟合的
从信号拟合的角度看, 式(7) 中理想的迭代学习信号
$u_{\mathrm{ILC}}^{*}=\left(S_{\mathrm{p}}\right)^{-1}\left(S r-S_{\mathrm{p}} F r\right) .$ | (30) |
4 结论针对光刻机工件台前馈控制,本文提出了一种基本前馈加额外有理分式补偿前馈的控制结构,减少了前馈控制中经典的四阶前馈拟合模型不准确对控制性能造成的影响。在此控制结构基础上,本文提出了一种基于数据驱动的额外补偿前馈的参数整定方法,通过优化广义残余重复误差将原非凸优化问题转化为凸优化问题,并给出了迭代过程中梯度、Hessian矩阵的无偏估计。通过光刻机工件台实验,证明这种前馈控制结构有效,并验证该参数整定方法具有可行性。相比于已有的基于ILC信号拟合的补偿前馈,本文研究方法在选用阶次较低补偿控制器时依然能够获得较好的轨迹跟踪控制效果。
参考文献
[1] | LI M, ZHU Y, YANG K M, et al. An integrated model-data-based zero-phase error tracking feedforward control strategy with application to an ultraprecision wafer stage[J]. IEEE Transactions on Industrial Electronics, 2017, 64(5): 4139-4149. DOI:10.1109/TIE.2016.2562606 |
[2] | LI M, ZHU Y, YANG K M, et al. Data-based switching feedforward control for repeating and varying tasks: With application to an ultraprecision wafer stage[J]. IEEE Transactions on Industrial Electronics, 2019, 66(11): 8670-8680. DOI:10.1109/TIE.2018.2886804 |
[3] | BAGGEN M, HEERTJES M, KAMIDI R. Data-based feed-forward control in MIMO motion systems[C]//Proceedings of 2008 American Control Conference. Seattle, USA: IEEE, 2008: 3011-3016. |
[4] | TOMIZUKA M. Zero phase error tracking algorithm for digital control[J]. Journal of Dynamic Systems, Measurement, and Control, 1987, 109(1): 65-68. DOI:10.1115/1.3143822 |
[5] | RIGNEY B P, PAO L Y, LAWRENCE D A. Nonminimum phase dynamic inversion for settle time applications[J]. IEEE Transactions on Control Systems Technology, 2009, 17(5): 989-1005. DOI:10.1109/TCST.2008.2002035 |
[6] | BUTTERWORTH J A, PAO L Y, ABRAMOVITCH D Y. Analysis and comparison of three discrete-time feedforward model-inverse control techniques for nonminimum-phase systems[J]. Mechatronics, 2012, 22(5): 577-587. DOI:10.1016/j.mechatronics.2011.12.006 |
[7] | DEVASIA S. Should model-based inverse inputs be used as feedforward under plant uncertainty?[J]. IEEE Transactions on Automatic Control, 2002, 47(11): 1865-1871. DOI:10.1109/TAC.2002.804478 |
[8] | HEERTJES M F. Data-based motion control of wafer scanners[J]. IFAC-PapersOnLine, 2016, 49(13): 1-12. DOI:10.1016/j.ifacol.2016.07.918 |
[9] | BRISTOW D A, THARAYIL M, ALLEYNE A G. A survey of iterative learning control[J]. IEEE Control Systems Magazine, 2006, 26(3): 96-114. DOI:10.1109/MCS.2006.1636313 |
[10] | BOEREN F, BAREJA A, KOK T, et al. Frequency-domain ILC approach for repeating and varying tasks: With application to semiconductor bonding equipment[J]. IEEE/ASME Transactions on Mechatronics, 2016, 21(6): 2716-2727. DOI:10.1109/TMECH.2016.2577139 |
[11] | VAN ZUNDERT J, BOLDER J, OOMEN T. Optimality and flexibility in iterative learning control for varying tasks[J]. Automatica, 2016, 67: 295-302. DOI:10.1016/j.automatica.2016.01.026 |
[12] | HUANG W C, YANG K M, ZHU Y, et al. Data-driven parameter tuning for rational feedforward controller: Achieving optimal estimation via instrumental variable[J]. IET Control Theory & Applications, 2021, 15(7): 937-948. |
[13] | JIANG Y, YANG K M, ZHU Y, et al. Optimal feedforward control with a parametric structure applied to a wafer stage[J]. Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering, 2014, 228(2): 97-106. DOI:10.1177/0954408913476442 |
[14] | VAN DER MEULEN S H, TOUSAIN R L, BOSGRA O H. Fixed structure feedforward controller design exploiting iterative trials: Application to a wafer stage and a desktop printer[J]. Journal of Dynamic Systems, Measurement, and Control, 2008, 130(5): 051006. DOI:10.1115/1.2957626 |
[15] | DAI L Y, LI X, ZHU Y, et al. Auto-tuning of model-based feedforward controller by feedback control signal in ultraprecision motion systems[J]. Mechanical Systems and Signal Processing, 2020, 142: 106764. DOI:10.1016/j.ymssp.2020.106764 |
[16] | BOLDER J, OOMEN T. Rational basis functions in iterative learning control-with experimental verification on a motion system[J]. IEEE Transactions on Control Systems Technology, 2015, 23(2): 722-729. DOI:10.1109/TCST.2014.2327578 |
[17] | BLANKEN L, BOEREN F, BRUIJNEN D, et al. Batch-to-batch rational feedforward control: From iterative learning to identification approaches, with application to a wafer stage[J]. IEEE/ASME Transactions on Mechatronics, 2017, 22(2): 826-837. DOI:10.1109/TMECH.2016.2625309 |
[18] | DAI L Y, LI X, ZHU Y, et al. Feedforward tuning by fitting iterative learning control signal for precision motion systems[J]. IEEE Transactions on Industrial Electronics, 2021, 68(9): 8412-8421. DOI:10.1109/TIE.2020.3020032 |
[19] | 戴渌爻. 超精密运动控制系统动态误差产生机理及控制方法研究[D]. 北京: 清华大学, 2021. DAI L Y. Research on the generation mechanism of dynamic errors and control strategies in ultraprecision motion control systems[D]. Beijing: Tsinghua University, 2021. (in Chinese) |
[20] | POTSAID B, WEN J T. High performance motion tracking control[C]//Proceedings of the 2004 IEEE International Conference on Control Applications. Taipei, China: IEEE, 2004: 718-723. |
[21] | HEERTJES M F, VAN DE MOLENGRAFT R M J G. Set-point variation in learning schemes with applications to wafer scanners[J]. Control Engineering Practice, 2009, 17(3): 345-356. DOI:10.1016/j.conengprac.2008.08.004 |
[22] | 黄伟才. 光刻机工件台数据驱动有理前馈控制器参数优化方法研究[D]. 北京: 清华大学, 2021. HUANG W C. Data-driven parameter optimization approach for the rational feedforward controller of the wafer stage of the lithography machine[D]. Beijing: Tsinghua University, 2021. (in Chinese) |
[23] | VAN ZUNDERT J, OOMEN T. On inversion-based approaches for feedforward and ILC[J]. Mechatronics, 2018, 50: 282-291. DOI:10.1016/j.mechatronics.2017.09.010 |