High Speed Localized Cooling using SiGe Superlattice Microrefrigerators zyx Yan Zhang, James Christofferson and Ali Shakouri Electrical Engineering Dept., University of California, Santa Cruz 1 156 High Street, Santa Cruz, CA 95064 Gehong Zeng and John E. Bowers Electrical and Computer Engineering, University of California, Santa Barbara, CA 93 I 06, USA Ed. Croke HRL Laboratories, LL Abstract In this paper, thin film based SiGe superlattice microrefrigerators are fabricated and characterized in terms of maximum cooling, power density and transient response. The localized cooling and fast transient response less than 40ps, demonstrate the potential for hot spot cooling of optoelectronics, microprocessor and IC chips. Keywords thin-film cooler, thermoelectric, thermionic superlattice, microrefrigerators, hot spots, optoelectronics, Introduction The current trend in optoelectronic and microelectric devices is to increase the level of integration with minimizing the die size, and at the same time increase clock speed (higher frequency). This results in higher power dissipation and an increase in the die temperature. According to the chip manufacture predictions, within the next five to ten years, the power requirements of the IC chip is going to exceed current cooling techniques [I], the case temperature need to be 20- 300C lower than its current values. Increasing device temperature causes problems for optoelectronics and IC chips. Typically temperature-dependent wavelength shifts for conventional laser sources are on the order of O.lnm/"C. Furthermore, according to electromigration model, the lifetime of IC chip is exponentially depend on its temperature, which could be represented by Black's equation [2]: zyxwvut T zyxwvutsrqponmlk = A/JZ*exp(Ea/kT) (eqn. zyxwvutsr 1) (K, mean time to failure, A, proportional constant, J, current density, Ea is the activation energy, typical value for silicon is about 0.68eV, k, Boltzmann's constant, T, absolute temperature) One distinguished characteristic of IC chips is uneven temperature distribution, leading to "hot spots". The temperature inside the chip could vary 5"C-30°C from one .location to another in microprocessor. For the case of optoelectronic devices, temperature difference between the active region and the heat sink area can be 100's of degrees. In terms of heat flux, current microprocessors have an average heat flux of 10-50 Wlcm', however, peak flux reaches as high as six times of its average value. [I] Thermal designs are driven by these hot spots instead of the whole chip temperature. Generally there are three alternative cooling technologies [I] with their own advantages and disadvantages: First, circulated liquid cooling, which could move heat sink away from processors therefore increase the surface area. However, it could not get effective heat sink resistance less 0-7803-7793-1/02/$17.00 zyxwvut 0 2003 IEEE 61 than 0.0 CIW; and reliability is also a concern if the liquid hose is leaking. Second method is refrigeration. The active cooling can provide an effective heat sink resistance less than 0.0 ClW. However, as the same problem as the first method, the limited space and reliability are the main concerns. The third methods are Thermoelectronic (TE) devices. Active cooling with no moving parts could provide an effective heat sink resistance less than 0.0 CIW. However, the low figure of merit, ZT, of current material impedes its industrial applications because of low efficiency. ZT=S20T/P, where S is the seebeck coefficient, zyxw o, electrical conductivity, 0. thermal conductivity and T, absolute temperature. Coefficient of performance (COP) of thermoelectric modules is directly related to ZT value. Typical commercial modules have a ZT-I, which corresponds to COP-0.6 for 30°C temperature difference. In addition, BiTelSbTe and PhTe, the common TE materials are both bulk technology, which is not compatible with standard microprocessor chips. The current smallest thermoelectric micro-modules have a short leg length on the order of 0.2-0.3mm, but with ceramic cap and thermal paste etc, the whole module is still near to 1 mm thick and 3- 4mm in diameter zyxwvu [3]. This is still too large for spot cooling. Another approach to eliminate hot spots is by theoretical simulation through optimized cell placement [4]. The temperature gradient inside the chip could be improved by a factor of two though at the cost of increasing wire length and cell area, which limits minimization and may bring more joule heating inside the chip. Through statistical methods of power and timing analysis, like McPower[5] and Mean Estimator of Density (MED)[6] etc., it is possible to find the nominal on- chip temperature profile. However these methods could not fundamentally remove the hot spots and reduce the IC temperature. Thus developing high cooling power density thin film refrigerators compatible with micro-fabrication process could have a strong impact in IC optimization. [7] Some recent exciting developments of thin film coolers and quantum dots devices have shown promising ZTs. For example, Rama Venkatasubramanian[8] demonstrated that the BiTe/ShTe superlattice could reach a ZT of 2.4 at 300K. Harman et. a1.[9] at MIT Lincoln lab demonstrated PbTe quantum dots with ZT of 1.6 -2.0 at 300K. All these could be developed for hot spot coolings. In our studies, we mainly focused on SiGe and InP materials, which are the substrate materials for microprocessor and optoelectronics. In previous studies, thin film coolers based on InP[ IO] and SiGe [I I] have been demonstrated. Devices fabricated on a conventional silicon substrate and diameter ranging from 150pm down to 20pm, have achieved 7 - 8 T cooling at IOOT ambient 19th IEEE SEMI-THERM Symposium