Category Archives: Science

科学,以及我的兴趣

推荐人工智能的书 (AI)

  1. AGENTS (Chap. 2): Agent and multiagent systems.
  2. SEARCH (Chap. 3-4): Heuristic search.
  3. GAMES (Chap. 5): Game theory and game search trees.
  4. CSP (Chap. 6): Constraint Satisfaction Problems.
  5. LOGIC (Chap. 7-9): Logical reasoning and representation.
  6. PLANNING (Chap. 10-11): Planning sequences of actions.
  7. UNCERTAINTY (Chap. 13-17): Representation and reasoning with probabilities.
  8. LEARNING (Chap. 18-21): Machine learning, staistics, inductive inference.
  9. NLP (Chap. 22-23): Natural language processing.
  10. VISION (Chap. 24): Computer and human vision (and other perceptual modalities).
  11. ROBOTICS (Chap. 25): Interacting with the physical world.
  12. PHILOSOPHY (Chap. 26): Philosophy and foundational works.
  13. NEUROSCIENCE (Chap. 26): The brain.
  14. PROGRAMMING: AI programming languages and techniques.

书单由顶尖高手Russell ,  Norvig 推荐

原文链接 http://aima.cs.berkeley.edu/books.html

Google的2011年校园招聘 笔试题

Google的2011年校园招聘宣讲会分别在北大和清华举行,其中北大本来是350人的会场,去了大约600多人,爆满,那场面绝对是人山人海,彩旗飘飘。经过了大约一个小时多的宣讲和问答,开始现场笔试环节,一共10个选择题和三个算法题,只有选择题答对了6个以上的人才有机会让面试官看你后面的算法题。然后明天下午会通知笔试通过的人进行面试,google的效率就像其搜索引擎一样迅速,效率可见一般。其中前10个选择题中有一个特别雷人的,题如下:现在北京有一套房子,价格200万,假设房价每年上涨10%,一个软件工程师每年固定能赚40万。如果他想买这套房子,不贷款,不涨工资,没有其他收入,每年不吃不喝不消费,那么他需要几年才能攒够钱买这套房子?

a, 5年

b, 7年

c, 8年

d, 9年

e, 永远买不起

我是这样解答的,先用数学工具做下抽象,本题可以简化为计算

n-5*1.1^n>0,简化步骤不在赘述,这个时候,可以用高中的作图法看焦点,或者简单观察算式,以可以得出结论。欢迎大家讨论跟好的解决方法。

模拟设计的100条圣经

1/ Capacitors and resistors have parasitic inductance, about 0.4nH for surface mount and 4nH for a leaded component.
2/ If you don”t want a high bandwidth transistor to oscillate place lossy components in at least 2 of the 3 leads. Ferrite beads work well.
3/ When taking DC measurements in a circuit and they don”t make sense, suspect that something is oscillating.
4/ Opamps will often oscillate when driving capacitive loads.
5/ The base-emitter voltage Vbe of a small signal transistor is about 0.65v and drops about 2mV/deg C. Vbe goes down with increasing temp.
6/ Multiply 0.13nV by the square root of the ohmic value of a resistor to find the noise in a 1Hz bandwidth. Then multiply by the square root of the BW in Hz gives the total noise voltage.
7/ Johnson noise current goes down with a increase in resistance.
8/ The impedance looking into the emitter of a transistor at room temp is 26Ohm/Ie in mA
9/ All amplifiers are differential in that they are referenced to ground somewhere.
10/ Typical metal film resistor has a temp coef of about 100 ppm/deg C
11/ The input noise voltage of a quiet op amp is 1nv/sqrt(Hz) but there are plenty available with 20nV/sqrt(Hz). Op amps with bipolar front-ends have lower voltage noise and higher current noise than those with FET front-ends
12/ Using an LC circuit as a power supply filter can actually multiply the power supply noise at the filter”s resonant frequency. Use inductor with low Q to overcome this.
13/ Use comparators for comparing and op amps for amplifying and don”t even think of mixing the two.
14/ Ceramic caps with any other dielectric other than NPO should only be used for bypass applications.
15/ An N-channel enhancement-mode FET needs +ve voltage on the gate-source to conduct form drain-source.
16/ Small signal JFETS work very well as low-leakage diodes by connecting drain & source together in log current-to-voltage converters and low leakage input protection. Small signal bipolars with b-c tied together will also make nice low-leakage diodes.
17/ With low pass filter use Bessel for least amount of overshoot in the time domain, and Cauer (or elliptic) for fastest rolloff in the freq domain.
18/ dB is always 10 times the log of the ratio of 2 powers.
19/ At low frequencies, the current in the collector of a transistor is in phase with the applied current at the base. At high frequencies the current at the collector lags by 90deg. You must appreciate this simple fact to understand high frequency oscillators.
20/ The most common glass-epoxy PCB material (FR4) has a dielectric constant of about 4.3 To make a trace with a characteristic impedance of 100 Ohm, use a trace thickness of about 0.4 times the thickness of the board with a ground plane on the opposite side. For a 50Ohm trace make it 2 times the thickness.
21/ If you need a programmable dynamic current source, find out about operational transconductance amps. Most of the problem is figuring out when you need a programmable dynamic current source.
22/ A CMOS output with an emitter follower can drive a 5V relay nicely as the relays normally have a must-make spec of 3.5V. This saves power and require no flyback components.
23/ Typical thermocouple potential is 30uV/degC. Route signals differentially, along the same path, avoid temp gradients. DPDT latching relays won”t heat up when multiplexing these signals.
24/ You SHOULD be bothered by a design that looks messy, cluttered or indirect. This uncomfortable feeling is one of the few indications that there”s a better way.
25/ Avoid drawing any current from the wiper of a potentiometer. The resistance of the wiper contact will cause problems (local heating, noise offsets etc.)
26/ Most digital phase detectors have a deadband where the analog output does not change over the small range where the 2 inputs are coincident. This often-ignored fact has helped to create some very noisy PLL”s (Use a high val bleeding resistor to always ensure current flow in the deadband)
27/ The phase noise of a phase-locked VCO will be at least 6dB worse than the phase noise of the divided reference for each octave between the comparison frequency and the VCO output frequency. Avoid low-comparison frequencies.
28/ You can almost always determine the leads of a bipolar transistor with an ohm meter. b-e and b-c junctions will measure like a diode with the b-c junction reading slightly lower than the b-e junction when forward biased.
29/ For low distortion, the drains (or collectors) of a differential amp”s front-end should be bootstrapped to the source (or emitter) so that the voltages on the part are not modulated by the input signal.
30/ If your design uses a $3 op amp, and you will be making a thousand of them, you have just spend $3000. Are you smart enough to figure out how to use a $.30 op amp instead?
31/ The Q of an LC tank circuit is dominated by the losses in the inductor in terms of series R. Q=omega.L/R
32/ Leakage current doubles for every 10degC increase in temp.
33/ When inputs to most JFET op amps exceed the common-mode range for the part, the output may reverse polarity. This artifact will haunt the designers of these parts for the rest of their lives, as it should!
34/ Understand the difference between “make-before-break” and “break-before-make” when you specify switches.
35/ 3 Terminal voltage regulators in the TO-220 packages are wonderful parts. They are cheap, rugged, thermally protected and very versatile. Use them virtually any place where you need a protected power transistor. They also make nice AM power-modulators.
36/ Use step recovery diode where you need fast edges under 100pS (hot-carrier is even faster)
37/ The old 723 regulator is still one of the lowest noise regulators around! (2.5uVrms 100Hz-10k)
38/ You can make a very simple oscillator with one diac, cap and a resistor.
39/ NPN transistors are normally superior to their PNP counterpart in performance.
40/ Typical spec in some databooks should read “Seen it once”. Always work with the worst spec of the part when doing a design.
41/ Don”t just copy circuits from application notes without understanding completely how it operates, and the reason for the choice of values.
42/ Dealing with crystals, make sure you understand the difference between series and parallel resonant. In a circuit, crystal frequency can generally be slightly lowered by placing a inductor in series and increased by a capacitor in series.
43/ Power MOSFETS on-resistance will have a -ve temp coef and not +ve at low current levels. This is important to remember when paralleling devices.
44/ Lowest noise figure of a RF transistor is not normally where the input is perfectly matched.
45/ Many un-stable RF devices can be made stable by loading the input or the output by a simple resistor, either in series or parallel.
46/ You trade gain for bandwidth.
47/ Push-pull power invertors using bipolars are risky and can saturate the core because of hysteresis stepping (use power fets)
48/ The Al value of a core will increase up to 50% or more under current transients.
49/ Be aware of leakage inductance when switching. V=L(dI/dt)
50/ The harder you turn-on a power transistor, the longer it will take to turn off.( the part where you burn the joules in the device)
51/ Always remember the Miller guy.
52/ In fault-finding a circuit, don”t overlook the obvious. (is there power?)
53/ What is a ground loop, and how to avoid it.
54/ 120 is a better number than 240 when using LM3XX type adjustable regulators.
55/ The lower comparator in the old 555 may have quite a long storage time.
56/ ZERO-ESR caps may do more harm than good.
57/ A correctly configured audio power amplifier will give more distortion in Class-AB, not less, because of the abrupt gain changes inherent in switching from A to B every cycle.
58/ Be a STAR when it comes to ground matters.
59/ Know when you need to use a Zobel network.
60/ Use current mirrors and mirror your current.
61/ Heatsink eff decreases with height above sealevel.
62/ A matt-black heatsink is much better than a shiny one.
63/ Ignoring secondary breakdown can be costly.
64/ Understand fuses and fuse ratings, fast and slow. Do you know when to use a semiconductor-fuse?
65/ Charge balancing resistors are a must when stacking serie-parallel high voltage capacitor banks.
66/ You must understand DC-restoration otherwise you will have a hard time designing Z-modulation in CRT circuits.
67/ Display 6 vert div low freq on a scope, increase the freq (make sure the source is constant amplitude) until display drops to 4.2 div. That is the true 3dB BW of the scope. (scope-source impedance should be matched)
68/ Doing a measurement with your DMM in the ACV position on your DC circuit will give a quick indication of any excess ripple on the supply when you don”t have a scope at hand.
69/ Dly timebase on a scope is very useful once you figured out when, why and how to use it.
70/ Know what to expect before you measure, otherwise any measurement is meaningless.
71/ Op amps. Output will swing in the direction that will force the inv-input level to try come closer to the non-inv input level.
72/ Understand virtual ground, slew-rate, CMRR and PSRR. (CMRR decrease with increase in freq)
73/ Making measurements near a spec-analalyzer”s noise floor will give 3dB errors.
74/ Understand the phase-noise limitations of the analyzer when making such measurements on oscillators.
75/ In a LC oscillator add some C with -ve temp coef to cancel the +ve temp coef of the L for min drift with temp.
76/ Less drift will result from making C with a few parallel caps, to reduce the heating effect of the oscillating current when spread out over a larger plate area.
77/ You will get more tuning range with the same LC combination in a Clapp than in a Colpitts circuit.
78/ High-Q tuned LC filters will have more insertion loss.
79/ Williams”s Rule (Guru at Linear Tech) for precision op amp circuits: ” Always invert (except when you can”t)”
80/ Cuk is not a kind of locomotive.
81/ If you don”t know how to make a design better, find out what makes it worse.
82/ Sometimes you know just enough to be dangerous.
83/ Impedance will reflect back as the square of the turns ratio.
84/ If you could design a component with the characteristics of a finger it could cure many design problems and you will be rich.
85/ Get nervous when the customer you are trying to help doesn”t even have a scope.
86/ Specs quoted by reps always exceed those by Engineering.
87/ A bad (Engineer) workman always blames his tools.
88/ Don”t believe everything that a SPICE program spits out.
89/ It is easy to get the color code of a 1kOhm and 12Ohm resistor mixed up when you are in a hurry.
90/ I bet one could write a thesis about the ability of probes to get tangled-up on a bench.
91/ DMM can upset sensitive circuits from noise generated inside it.
92/ When probing directly on a crystal of a uP, use 10kOhm or so resistor in series with the probe tip to prevent loading from stopping the osc.
93/ It is easier to see what is happening on the ports using a scope when you trigger one chan against the cpu clock.
94/ National once made a bad op amp many years ago that some Engineers referred to it as “Jelly Beans”
95/ The moment you can start to notice distortion on an oscilloscope it is already way past being acceptable.
96/ Be big enough to say “I don”t know”, people will respect you more.
97/ The best designer is often working in the marketing department.
98/ Some remarkable discoveries/inventions were made by people that knew very little about the subject. Don”t fall into a groove in you thinking process.
99/ The Peter-Principle : Everybody will be promoted up to his own level of incompetence. http://pespmc1.vub.ac.be/PETERPR.html
100/ END-Enjoy

解密3项关键发明,多晶硅太阳能电池变身大众能源不是梦!

麻省理工学院(MIT)科学家近日公布了可大幅提高多晶硅太阳能电池效率、同时维持低成本的方法。与此同时,这些科学家们还成立了一家名为1366的技术公司以将这项技术商业化。伊曼纽尔.萨克斯是MIT机械工程学教授,也是1366公司的创办人之一。他的实验室研制出的大约2厘米宽的小型多晶硅太阳能电池,其光电转换效率比普通多晶硅太阳能电池提高了27%。

据悉,萨克斯采用了3项关键的发明来提高太阳能电池模型的效率。

第一,在太阳能电池表面增加纹理,使硅板能吸收更多的光。当光线进入电池时,粗糙的表面使得光线发生弯曲,当光线到达电池的背面时,它不会被直接反射出去,而是被小角度反弹回,从而驻留在硅太阳能板里。光线在硅板里停留的时间越长,它被吸收而转换成电能的几率就越大。这项技术曾在单晶硅太阳能电池上使用,但此前在多晶硅电池上还很难实现。

第二,萨克斯发明了一项技术,可以制备很细的银丝——其直径只有太阳能电池通常使用的银丝的五分之一,而且提高了导电率。银丝越细,制造成本就越低。同普通银丝相比,细银丝可以更紧密的排列,彼此的间隔更小,这使得银丝采集电流的效率更高。

第三,使用一套宽平的金属条来汇集通过细银丝传来的电流。通常,这些金属条会阻碍光线进入太阳能电池,从而使电池效率下降。但萨克斯通过蚀刻金属条表面,使其变得像多面镜一样,从而获得了与在硅板表面增加纹理一样的效果。虽然这道工艺步骤会使生产成本增加,但银的用量减少了,二者可以抵消。

1366公司正在兴建一座试验工厂,以生产完全尺寸的太阳能电池(大约15厘米宽)。萨克斯表示,如果公司的放大生产能够成功,这项技术将使太阳能发电的成本大幅降低。他估计,第一批采用新技术生产的太阳能电池,其发电成本约为1.65美元/瓦,考虑到今后的技术改进,成本会降为1.30美元/瓦。但是要想和煤炭发电竞争,太阳能发电的成本必须降到1美元/瓦才行。萨克斯预测,通过改进减反射涂层和其它技术进展,在2012年左右,这个目标完全可以达到。